What's new

Google’s new version of its video model generates clips with native audio and can extend an existing video.

It works from text, images, and video; references keep characters and objects consistent, per Google, and a reference video’s audio is ignored.

• Output: 3 to 15 seconds per generation, at 360p, 720p, 1080p, or 4K.
• References: up to 6 images and up to 3 short videos, used for appearance only.
• Frames: a required start frame and an optional end frame.
• Extend: continues a video from its end, 3 to 10 seconds per step, up to a total of 30 seconds.

To get started: Video → Gemini Omni Flash 1.2 → choose to generate with References, Frames, or Extend.

Availability: on the web, in Higgsfield MCP, Supercomputer, and CLI.

OpenAI’s new image model is now on Higgsfield in two variants: Flare and Sunburst.

Besides standard images, it makes animated GIFs, vector SVGs, and generates on a transparent background.

• Flare: the default variant, for everyday content, drafts, and series of images.
• Sunburst: for detailed edits and final layouts where faces, products, and text need to stay close to the source; generation takes longer.
• Settings: quality from Low to Max, 1K, 2K, or 4K, opaque or transparent background, and PNG, GIF, or SVG output.
• GIF: a scene and one simple movement become a loop; frame count depends on frame size, from 2 frames at 1024 px to 64 at 288 px.

To get started: Image → GPT Image 2.6 Flare or Sunburst → Generate.

A lighter version of Kling 4.0 for quick tests and frequent short videos.

It covers the same modes, References, Frames, and Edit video, with shorter clips and fewer controls; for final versions with keyframes and 1080p, use Kling 4.0.

• Length and resolution: 3 to 20 seconds at 720p.
• Frames: a start frame only, which also sets the aspect ratio; no last frame or keyframes.
• References: up to 10 images and 5 videos, plus Elements with @ in the prompt.
• Cost: lower per generation than Kling 4.0; the exact price is on the Generate button.

To get started: Video → Kling 4.0 Flash → choose to generate with References, Frames, or Edit video.

A tool for vertical 9:16 reels with an AI avatar as the host: add a character, give a topic, and upload a video that sets the motion.

The script is written for you from a topic, a hook, or a few key points. Motion comes from a reference video of 10 to 120 seconds or from a ready format in Explore.

• Character: add an image or a video from your Uploads, AI Influencer characters, and Generations.
• Duration: 10 seconds to 2 minutes on a slider; the credit cost changes with length.
• Quality: Basic or High.
• Recreate: any format from Explore can be repeated with your own character.

To get started: open Shorts → add a reference video and a character → type your topic → Generate, on Higgsfield or in Claude through Higgsfield MCP.

Generations on all Kling models, including Kling 4.0, Kling 3.0, and Kling Motion Control, were failing. The cause was a temporary provider-side issue. It's resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

A new tool for manga, comics, and webtoons: start with an idea or a script, create characters, generate chapters and pages, edit the result, and export the book.

Recreate reimagines an existing manga with your own cast: replace the characters with photos, change the title, and add a prompt for other changes.

• Start: describe an idea, or upload a script with references. Choose a style (B&W manga, color manga, comics, or webtoon), an aspect ratio, and one of 14 languages.
• Modes: Auto creates characters, locations, props, and chapters on its own; Manual lets you generate and check each one.
• Size: 4 to 30 chapters of about 20, 30, or 40 pages.
• Export: download the whole book, single chapters, or just the cast and world in PDF, CBZ, PNG pages, or webtoon format. A finished book can also be published on Higgsfield.

To get started: open Mangaka → Create from idea or Upload your references → set style, format, and language → Generate.

The winners of the $1,000,000 festival were revealed in a YouTube premiere, and are listed on the festival page.

An independent jury of four filmmakers, Edwin Catmull, Phedon Papamichael, Paul W. S. Anderson, and Alex Proyas, chose 1st to 3rd place and 10 Honorable Mentions; Audience Choice went to the film with the most valid votes from paid-plan users.

• Criteria: Directing & Storytelling, Visual Craft, Sound & Editing, and Originality & Concept.
• Prizes: 1st $500,000, 2nd $200,000, 3rd $100,000, Audience Choice $100,000, and 10 Honorable Mentions of $10,000 each. One prize per film.
• Higgsfield Talent: a separate creator support program launched alongside the results, with filmmakers selected by the Higgsfield creative team.

To get started: see the winners on the festival page.

A tool for talking characters: pick a character and a location, add speech, and get a lip-synced video with a voice that stays the same from scene to scene.

Speech is limited to 30 seconds per generation.

• Character: an uploaded image, including your own photo, or one from AI Influencer.
• Speech: an uploaded recording or newly generated audio. Cues in square brackets, such as [smirks] or [short pause], set the delivery.
• Settings: 720p or 1080p, five aspect ratios, and an optional prompt for movement and camera.

To get started: Video → Higgsfield Intone → Add character, location, and audio → Generate.

Ideogram’s model for targeted image edits is now on Higgsfield: only the selected part of a picture changes, at its original resolution.

Two modes: Whole image edits the entire picture and scales large files down to about 2K; Select area edits only the framed part and returns the result at the size of the upload. Works with images up to 8K.

• Select area limits: each side at least 256 px, total area up to 4.2 MP, aspect ratio up to 6:1.
• Quality: Very low, Low, Medium, or High.
• Results: 1 to 4 per run.

To get started: Image → Ideogram 4.5 Precise Edit → upload an image → Select area and describe the edit.

OpenAI’s fast, low-cost model with visual analysis and computer use is now in Supercomputer on Higgsfield.

It’s built for everyday work: writing, reading images and documents, and multi-step tasks in apps. For the hardest coding, OpenAI recommends GPT-6.1 Sol or GPT-6 Astra.

Per OpenAI:

• Speed: completes tasks more than 3x faster than GPT-6 Astra and GPT-6.1 Sol.
• Accuracy: fewer factual errors and hallucinations than GPT-5.6 Terra.

To get started: open Supercomputer and select GPT-6.1 Terra.

Build a mobile game from a description or a gameplay video and play it in the browser within minutes; for Enterprise, the same engine makes playable ads.

Playable builds the game itself and is available to everyone; Playable Ads is an Enterprise product for interactive ads of games, with access through Contact Sales.

• Input: a text description, or a gameplay video the tool reads for level layout, controls, and timing.
• Settings: 8 genres from Shooter to Simulation, 2D, 3D, or 3D Fast, and Manual or Auto mode.
• Edits without code: difficulty, player, enemies, and camera via a slider or a text request; the hero or palette can change while the mechanics stay the same.
• Playable Ads: optional assets (up to 40 images, one video, up to 20 audio files, up to 3 playables) and export as one HTML file up to 5 MB for Generic MRAID, Moloco, AppLovin, Google App Campaigns, Unity Ads, Meta, Pangle, Mintegral, or Liftoff.

To get started: open Playable → describe the game or upload a video → preview → Create game. For Playable Ads, use Contact Sales on Pricing.

Availability: Playable on the web, on mobile, and in Claude through Higgsfield MCP; Playable Ads on Enterprise plans.

Ideogram’s model for vector graphics is now on Higgsfield: SVG logos, icons, and illustrations from a prompt or an existing image.

Vector output stays sharp at any size, which suits print and scaling.

• Text to SVG: a prompt with optional references, with aspect ratio and Low, Medium, or High quality settings.
• Image to SVG: one image converted into editable vector shapes; no prompt needed. Simple logos and illustrations work best, detailed images get simplified.

To get started: Image → Ideogram Vector 1.0 → select Text to SVG or Image to SVG.

Moonshot AI’s new open-weight model is now in Supercomputer on Higgsfield.

It’s one of the lower-cost models in the selector, built for long tasks.

• Context: up to 1M tokens, with text and image input.
• Best for: whole-project code, 3D modeling in Blender and CAD, and financial calculations, per Moonshot AI.

To get started: open Supercomputer → model selector → Kimi K3.1 Pro.

A new tool that turns a “what if” topic into a finished clip: script, characters, scenes with voiceover, and edit.

• Length and batches: videos from 15 seconds to 5 minutes, 1 to 4 episodes per run.
• Presets: a library of 50+ formats, or your own built from links to existing videos.
• Voiceover: each line is generated separately and can be redone on its own.

To get started: open Content Factory → pick a format → enter a topic in What if → Generate.

The weekly cinematic challenge moved from our Discord Community to the platform: brief, submissions, and results on one page, with films submitted from Cinema Studio.

Any Higgsfield account can enter, free or paid; Discord is optional and hosts the Wednesday stream where winners are announced.

• Format: a film of exactly 120 seconds, horizontal 16:9 or 21:9, with all AI generation inside the challenge project. Editing in other tools is allowed.
• Schedule: a new brief every week with one technical focus; submissions close Tuesday at 17:00 UTC, winners on Wednesday at 17:00 UTC.
• Prizes: 30,000, 20,000, and 10,000 credits plus honorable mentions. Credits arrive within 7 days and are valid for 90 days.
• First round: WCC#1 Impossible Camera, October 21 to 28, 2026: a story told in one continuous camera move.

To get started: open Contests → Weekly Challenge → Start your project → make the film in Cinema Studio → Submit on the project page.

Kling’s new video model is now on Higgsfield. It generates from text, an image, or a set of references and can edit finished clips.

A launch discount is applied for a limited time, as of October 21, 2026; the exact price is on the Generate button.

Compared with Kling 3.0, per Kling:

• Motion: more stable in fast scenes such as fights, sports, and character interaction, with stronger camera work.
• Keyframes: up to 10 across the video instead of first and last frame only.
• References: up to 15, including images, up to 5 videos, and Elements; Creative Recreation reuses the pace, composition, and motion of an existing video.
• Video Editing: swap a character, product, background, camera angle, or style in a finished clip.
• Output: 3 to 30 seconds, 720p or 1080p, aspect ratios from 21:9 to 9:16, stereo sound, and voiceover in 9 languages.

To get started: Video → Kling 4.0 → References, Frames, or Edit video.

Higgsfield is now available in Chinese for all users, with no third-party translation needed.

Signed-in users can switch languages from the profile menu or Supercomputer; signed-out users can use the language selector in the footer.

A lighter version of Genjutsu for short clips, with a lower credit cost.

Motion transfer and Objects swap work the same way as in Genjutsu; for longer videos, more references, or Restyle, switch to the standard version.

• Inputs: a video from 4 to 15 seconds and up to 10 reference images.
• Modes: Motion transfer and Objects swap.
• Quality: 720p or 1080p.

To get started: Video → Higgsfield Genjutsu Lite → pick a mode → upload a video and references → Generate.

The Higgsfield plugin in ChatGPT now turns a photo, video, or text description into a 3D scene you can reshoot from new angles.

For better results, select GPT-6 Astra before starting building your scene.

• Camera: lenses from 24 to 300 mm, apertures from ƒ1.4 to ƒ16, and eight frame formats from 1:1 to 21:9.
• Controls: move with W, A, S, D, rotate by dragging, zoom with the scroll wheel.
• Output: a photo or a video generation of up to 30 seconds; draft videos let you check the camera move before the final version.

To get started: open the Higgsfield plugin → 3D Angles → Upload a reference, or Describe the scene.

A $50,000 Discord contest for content made on Higgsfield with investor and advisor Iman Gadzhi in the lead role: ten prizes of $5,000.

Five prizes go to the posts with the strongest Instagram engagement, five to the most creative entries picked by Iman and the team. Rules and the submission are in Discord.

• Task: a short film, ad, trailer, poster, or meme about his investment, made on Higgsfield in one project.
• Entry: post it on Instagram with #HiggsfieldxIman, tagging @higgsfield.ai and @imangadzhi, then submit under the announcement in #updates.
• Dates: entries close October 24, 2026; results will be announced in Discord.

To get started: join the Higgsfield Discord → link your account in #verification → Submit.

Terms: 18+, solo or as a team. Prizes are paid in money, not credits.

A beginner course on Higgsfield that goes from a first video to directing a scene, with a certificate on completion.

Watching is free; the required homework needs an active subscription and uses credits from your balance.

• Lessons: an introduction and 7 lessons, 1 h 42 min in total, each with materials in PDF.
• Homework: your own generations from the last 3 days; required tasks unlock the next lesson, with no grading.
• Rewards: a certificate shareable to LinkedIn, the Foundation achievement, and a Creator Drop.

To get started: open Academy → Courses → AI Creation: First Steps.

A new build of the upscaler is live on Higgsfield. It replaced the previous version in Upscale automatically, so there’s nothing to switch or set up.

• Quality: sharper detail, with motion and audio still preserved, per Black Forest Labs.
• Speed and cost: shorter processing time and fewer credits per upscale; exact rates are shown before generation.

The updated version is currently available only on Higgsfield.

To get started: Upscale → FLUX 3 Video Upscaler → upload a video.

A public page with a curated feed of creators’ shots and open projects, visible without an account.

• Tabs: Discovery with Featured Creators, Projects, Shots, Creators, Contests, and Originals.
• Getting featured: a published work with 30 likes goes to the team for review, or you can submit an Open Project through Higgsfield Grants.
• Grants: up to 200,000 credits for an Open Project with a video of at least 1 minute.

To get started: open the Community page, or submit a project on the Grants page.

A free week of Wan 3.0 and Wan 3.0 Prime for subscribers, and 60% off the same model for developers until October 31, 2026.

Terms as of October 13, 2026.

• Limits: clips up to 30 seconds at up to 1080p, in a shared queue, so speed depends on load.
• Upgrades: moving from Basic or Starter to a higher tier during the offer starts your 7 days from that moment.
• Discount: applies to all API users; the standard rate returns after the offer ends.

To get started: Video → model selector → Wan 3.0. For developers, rates are on Higgsfield API.

Availability: individual plans from Pro and above, monthly or annual. Current plans are on Pricing.

A new Higgsfield skill for Claude that makes video edits in Aura Farming, Motion, or Clipping styles from one reference photo.

It runs on Higgsfield models through MCP: pick a preset on the Katana page or in a Claude chat, add your references, and the skill generates the edit.

• Presets: a Higgsfield library in three categories, available on the site and in Claude.
• References: up to 5 images of characters, products, or clothing per edit. To adjust the result, switch on Prompt and describe the change.
• Community presets: share your own from the Katana page with a 3 to 30 second video for the motion, up to 40 images, and a prompt that mentions them with @. Shared presets go live right away and can be taken down after reports.

To get started: open the Katana page → pick a preset → upload your images → Recreate in Claude. Or ask Claude directly: “make me an edit of this product”.

Vote for shortlisted films until October 15, 2026, 11:59 PM UTC. The film with the most votes wins the $100,000 prize.

• Shortlists: Create in Public, the top projects by community reactions (up to 50), and Higgsfield Picks, selected by the team. Likes from Create in Public don’t carry over; voting starts from zero.
• Votes: up to 5 per user, one per film, after watching it all the way through. A vote can be moved to another film anytime until voting closes.
• Prizes: Audience Choice is separate from the jury prizes for 1st to 3rd place and honorable mentions. One film can win only one prize.

To get started: open the festival page → watch a shortlisted film → Vote.

Availability: accounts with a paid plan. You can’t vote for your own film.

A community contest to celebrate reaching 200,000 members on our Discord server.

Take any viral video trend and make your own version of it on Higgsfield. The prize pool is 200,000 credits, and the community picks the winners by voting. Full rules and FAQ are in Discord.

• Entry: a video up to 60 seconds, made in your contest project in Cinema Studio. One entry per person.
• Prizes: 1st 50,000 credits, 2nd 32,000, 3rd 20,000, 4th to 10th 14,000 each. Promotional credits, valid for 90 days.
• Dates: entries close October 14, 2026, 17:00 UTC. Voting runs October 15 to 17 on the site. Winners are announced October 19 in Discord.
• Public project: once you submit, your project, prompt, and settings are visible to everyone.

To get started: join the Higgsfield Discord → apply to create your contest project → make your video → Submit.

Availability: 18+, with a Higgsfield account linked to Discord. Voting is open to accounts with an active paid plan.

Recreate any video ad with your own characters, products, and locations from one reference, now with a visual interface on Higgsfield.

Upload an ad, add your assets, and describe the new scene. Ad Multiplier keeps the original's structure and swaps in what you specify.

• What you can change: product, cast, location, script, hook, or CTA. Mention attached files in the prompt with @.
• Hooks and CTAs: try as many as you want without re-uploading the reference.
• References: one video from 4 to 30 seconds and up to 20 images.
• Quality: 480p, 720p, or 1080p.

To get started: open Ads Studio → Ad Multiplier → upload a reference video → Generate.

Google's update to Nano Banana 2 is now available on Higgsfield, with the same speed and output up to 4K.

Compared with Nano Banana 2, it improves visual quality, prompt adherence, character consistency across edits, and text rendering, per Google.

• References: up to 14 images, with consistency for up to 4 characters and up to 10 objects.
• Text edits: describe what to change, such as a shirt color, and the model edits that part of the image.
• Wide formats: fixed tiling artifacts on 1:4, 4:1, 1:8, and 8:1 at 2K and 4K.
• Resolution: 1K by default, 2K, or 4K.

To get started: Image → model selector → Nano Banana 2.1.

The festival timeline has been extended to allow more time to review all submissions.

• Shortlist and Audience Choice voting: October 8, 2026.
• Jury review: October 9–14.
• Winners: announced in the last week of October.

See the current dates on the Festival Timeline.

A provider-side issue caused failed and slow generations, mostly on 2.0, 2.0 Mini, 2.0 Fast, and 2.5 Draft Mode.

All models are running normally again. Credits for failed generations were refunded automatically, so you can run them again at no extra cost. Sorry for the disruption.

Build your own AI character, then bring it into videos on Higgsfield.

• Starting point: create a character from scratch, from a reference, or from your own photo with Add your face.
• Character Builder: 19 categories, from character type and age to style and accessories. Shuffle picks a random set.
• Videos: click Use in Genjutsu to put your character into an existing video with Motion transfer or Object swap.
• Free generations: 5 for every user to try it, as of October 3, 2026.

To get started: open AI Influencer → Character Builder → Generate.

Earn a $15 bonus after your first $100 of API spend, plus up to 20% cashback in 30-day cycles based on how much you spend.

Cashback is earned on API usage paid from your main balance, not on top-ups. Generations paid with cashback don't count toward your spend.

• $15 bonus: credited after you spend your first $100 from your topped-up balance.
• Cashback levels: your API spend over 30 days sets your cashback for the next 30 days: $5,000 → 5%, $10,000 → 8%, $20,000 → 12%, $50,000 → 20%.
• Cycles: your first 30 days start on your sign-up date, and everyone starts at 0%.
• Paying with cashback: your cashback balance covers a generation when it has enough for the full cost. Otherwise, your main balance is charged.

To get started: track your progress and current level on the cashback page.

xAI's new video model, with sound generated in the same pass as motion, is now available on Higgsfield.

• Modes: image-to-video, text-to-video, reference-to-video, and video extension.
• Video extension: continue a clip from its last frame, keeping the character, lighting, and motion consistent.
• References: up to 7 images to keep a character or style consistent across generations.
• Output: 480p, 720p, or 1080p; 1 to 15 seconds; 24 fps.

To get started: Video → model selector → Grok Imagine Video 1.5.

A new model for creating and editing images is now available on Higgsfield.

It's part of the multimodal FLUX 3 family. Compared with earlier versions, it follows complex prompts more closely, renders text more accurately in multiple languages, and makes more precise edits, per Black Forest Labs.

• Modes: text-to-image and image-to-image.
• Styles: photo, illustration, poster, collage, and editorial.
• Resolution: 768p, 1K, 1.5K, 2K, or 4K, with native rendering at 2K and 4K.

To get started: Image → model selector → FLUX 3 Image.

Seedance 2.0, Seedance 2.5, and Seedance 2.5 Draft Mode generations were failing or running slow. The cause was a temporary provider-side issue. It's resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

Turn a link to your website into static ad creatives with your brand, products, and design on Higgsfield.

• Brand setup: set your logo, colors, tone of voice, audience, and selling points once, then reuse them in every batch.
• Products: pick products from your site, or add them by link or manually.
• Batches: up to 20 ads per product at a time.
• Chat: use Mention a product to generate more ads for a product in your project.

To get started: open Ads Studio.

A new image model built for editing is now available on Higgsfield.

Edit the same image repeatedly while colors, textures, and details stay in place, per Ideogram.

• Reference edits: transfer a style or element from up to 4 reference images.
• Text edits: replace or translate text in a design with a prompt, keeping the layout.
• Transparent background: add "transparent background" to your prompt to generate without one.
• Resolution: 1K or 2K.

To get started: Image → model selector → Ideogram 4.5.

Redraw a whole video in a new visual style on Higgsfield, keeping the original motion, timing, and camera work.

It works in Genjutsu's Motion transfer mode, with reference videos from 4 to 30 seconds.

• Styles: 20+ animation styles, or build your own from reference images.
• References: up to 30 images of characters, products, or clothing.
• Resolution: 480p, 720p, or 1080p.

To get started: Video → Higgsfield Genjutsu → Styles.

Availability: on the web and via MCP.

Generate images, video, branded content, and websites right in ChatGPT, built on OpenAI's new extensions feature.

• Slash commands: /effects for Viral Presets, /product, /motion, and /marketing-studio for Marketing Studio, and /genjutsu for Genjutsu.
• Ads and UGC: product ads, video ads, UGC videos, and Ad Multiplier variations from one photo or video.
• Sites and games: build, publish, and update websites, web apps, and browser games in the chat.

To get started: install the Higgsfield plugin.

Generations failed more often than usual on 2.0 and took longer to complete on 2.5 because of a provider-side issue.

Both models are running normally again. If any of your generations failed, you can run them again. Sorry for the disruption.

Always-on AI agents in ChatGPT can now generate video and images on Higgsfield through the Higgsfield plugin.

Each dot runs on GPT-6.1 Sol with its own cloud computer, per OpenAI.

• Higgsfield tools: Cinema Studio, Genjutsu, Marketing Studio, and more.
• Chat: ChatGPT, Slack, and Teams. Set up your first dot on desktop.

To get started: install the Higgsfield plugin.

Availability: ChatGPT Pro and Business Premium in supported regions, plus Enterprise, Edu, and Healthcare when an admin turns on Dots.

GPT-6.1 Sol, OpenAI's upgrade to GPT-6 Sol, is now available in Supercomputer on Higgsfield.

It's a text model for code and agentic tasks, and Supercomputer uses Orchestrator by default, so select GPT-6.1 Sol manually to use it.

Compared with GPT-6 Sol, per OpenAI:

• Coding: better in real codebases, close to GPT-6 Astra on long engineering tasks.
• Documents: more accurate answers on complex PDFs with tables, charts, diagrams, and small print.
• Computer use: works better with desktop apps, with results close to GPT-6 Astra.
• Science: stronger at data analysis, simulations, and terminal work. For the hardest research tasks, OpenAI still recommends GPT-6 Astra.

To get started: open Supercomputer → model selector → GPT-6.1 Sol.

Subscribers on Higgsfield get their credit balance topped up to their plan's full monthly amount.

The offer ends September 30, 2026, at 23:59 UTC. Unused bonus credits expire at that time.

• Eligibility: Paid plans from Basic and above. Upgrades don't add bonus credits.
• Bonus credits: Shown separately in your credit balance and spent first.
• Canceled plans: If your plan is still active, you can resume it and get the reset on the same terms.

To get started: claim your bonus credits in Manage Account → Subscription.

Eleven v4, ElevenLabs' new text-to-speech model, is now available in Audio on Higgsfield.

It adapts delivery to the text, from dramatic to soft, urgent, or humorous, while keeping a consistent voice, per ElevenLabs.

• Multiple speakers: several voices in one generation, for dialogues, podcasts, and scenes.
• Delivery control: follows tags and delivery prompts more closely than earlier ElevenLabs models, per ElevenLabs.
• Languages: 90+, with the native accent held for the whole generation.
• Length: up to 10,000 characters per generation.

To get started: open Audio → model selector → Eleven v4.

Claude Sonnet 5.5, the new version of Anthropic's Sonnet model, is now available in Supercomputer on Higgsfield.

• Speed: more than 30% faster output than Claude Sonnet 5, with fewer tokens for the same work, per Anthropic.
• Coding: stronger than Claude Sonnet 5 at agentic coding, including building websites and apps and fixing bugs, per Anthropic.
• Best for: well-defined tasks like docs, slides, spreadsheets, and quick iterations. For complex, open-ended work, use Claude Opus 5.5.

To get started: open Supercomputer → model selector → Claude Sonnet 5.5.

Higgsfield on mobile now opens on the prompt bar, with navigation moved to a sidebar.

Features are unchanged; only their placement has moved.

• Agent tab: popular scenarios, including Genjutsu, GPT-6 Astra, Create Image, and Create Video.
• Sidebar: opens from the top-left icon, with New Chat, Explore, MCP, Higgsfield Genjutsu, Community, and Library.
• Recent Chats: below the main sidebar items.
• Profile: at the bottom of the sidebar, with your credit balance and account details.

Higgsfield API returns $1 in cashback for every $1 spent on API usage, valid until September 30, 23:59 UTC.

Cashback is earned on API usage paid from your main balance, not on top-ups. The campaign ends early if the $20M pool runs out, and unused cashback expires at the deadline.

• Limit: up to $1,000 per account; verify a business email and add a card to raise it to $200,000.
• How it's spent: cashback pays first when it covers a generation's full cost; otherwise your main balance pays and earns cashback.
• Discounts: generations paid with cashback use standard prices, without model discounts.

To get started: top up in the API console and use the API as usual.

The Production Skills Bundle adds 11 Higgsfield workflows for Claude MCP that run inside Blender, Adobe apps, TouchDesigner, and DaVinci Resolve Studio.

The agent works in your installed software and returns an editable project. Claude Desktop and the software need to be on the same computer.

• 3D (Blender): Destruction Studio, Exploded View, Scene Builder, Cartoon Shaders.
• Editing and VFX (Premiere Pro, After Effects): Project Sorter, Shot Composer, Shot Cleanup.
• Design (Illustrator, Photoshop): Vectorize, Image Fixer.
• Effects and color (TouchDesigner, DaVinci Resolve Studio): TouchDesigner Effects, Color Grading.

To get started: connect Higgsfield MCP in Claude Desktop → open the Production Skills Bundle → Try in Claude.

A new image model from ByteDance is now on Higgsfield, with multi-reference style blending and editing in the same model.

It's the newest model in the Seedream line, built for speed and visual reasoning, per ByteDance.

• Resolution: 1K, 1.5K, or 2K.
• Multi-reference: add several references and the model blends their style in one frame.
• Generation and editing: create from scratch or edit an existing image with the same model.

To get started: Image → model selector → Seedream 5.0 Flash.

Anthropic's new Opus model is now available in Supercomputer on Higgsfield.
Compared with Claude Opus 5, per Anthropic:

• Efficiency: completes typical tasks with fewer tokens.
• Coding and knowledge work: better at agentic coding, long multi-step tasks, reports, analysis, and financial models.
• Output: reports what it did, what it found, and what it needs next.
• Visual input: reads charts, diagrams, and screenshots more accurately. Computer use at default effort matches Opus 5 on high.

To get started: open Supercomputer → model selector → Claude Opus 5.5.

A two-step mode on Higgsfield: a 480p draft first, then a 1080p render of the one you choose.

A draft costs fewer credits than a full-quality video, so testing several versions of one idea is cheaper. The final render keeps the original settings: prompt, references, seed, aspect ratio, and duration.

• Iterations: generate as many previews as you need, changing the text or images between them.
• First and last frame: upload a start image and an end image, and the model fills in the motion between them.
• Sound: voices, effects, and background noise are generated with the video, so you can check them before upgrading.
• Length: up to 30 seconds.

To get started: Video → Seedance 2.5 Draft Mode → generate → select a preview → generate in 1080p.

Availability: all regions except the US, where the provider doesn't offer the model.

OpenAI's new image model is now on Higgsfield in two variants: Flare for most tasks and Sunburst for precise edits.

Compared with GPT Image 2, per OpenAI:

• GPT Image 2.5 Flare: higher quality at up to 50% lower latency.
• GPT Image 2.5 Sunburst: precise control in edits; generation takes longer.
• Both: sharper detail, better subject preservation from references, targeted edits that hold up across iterations, and stronger style holding.

To get started: Image → GPT Image 2.5 Flare or GPT Image 2.5 Sunburst.

OpenAI's two new GPT-6 models are now in Supercomputer and Higgsfield MCP. They are trained with the same methods as GPT-6 Astra but tuned for price and speed; Astra stays the flagship.

Per OpenAI:

• GPT-6 Sol: the workhorse of the line, for complex coding, agentic tasks, long multi-step scenarios, and tool use.
• GPT-6 Luna: the cheapest model in the family, for simple, high-volume tasks.

To get started: open Supercomputer and select GPT-6 Sol or GPT-6 Luna, or install the Higgsfield plugin for ChatGPT.

xAI's new model for coding and knowledge work is now in Supercomputer on Higgsfield.

Compared with Grok 4.6, per xAI:

• Size and context: larger, trained longer on multi-hour tasks, and holds longer context.
• Checking: reviews its own output more carefully.
• Everyday work: tuned for conversation as well as code, with fewer slips on routine tasks, documents, and presentations.

To get started: open Supercomputer and select Grok 4.7.

A standalone product for developers: one key opens 50+ models for image, video, and audio. You send a prompt; Higgsfield handles the queue, GPUs, and scaling, and returns a link to the file.

Billing is separate from higgsfield.ai plans and credits.

• Pay per generation: no subscriptions or seats; public per-model rates, and only successful generations are billed.
• Three calls: submit a prompt, poll the status, fetch the file.
• SDKs: Python and TypeScript, or plain REST.

To get started: create a key at Higgsfield API.

Seven motion workflows for After Effects run on GPT-6 Astra through the Higgsfield plugin for ChatGPT, each with a ready prompt, a workflow breakdown, and a Try in ChatGPT button.

• Presets: Illustration Animation, Localization Motion, Motion Design, Paper Collage, Presentation Animation, SaaS Animation, and Whiteboard Animation.
• Bundle: all seven together as the Motion Design Bundle.

To get started: open the ChatGPT plugin page → pick a preset.

An agent that connects Higgsfield to Adobe After Effects through the Higgsfield plugin for ChatGPT. Generate visuals in Higgsfield, describe the motion work, and the agent builds it in your open composition: titles, logo reveals, transitions, promos, and effects.

It requires active Higgsfield, ChatGPT, and After Effects subscriptions; generations use your regular credits.

• Input: your material and references, translated into layers, keyframes, and timing.
• Settings: resolution from 1080p to 8K and duration, set in the prompt.
• Output: native After Effects elements, editable as usual.

To get started: install the Higgsfield plugin for ChatGPT and type /use-after-effects.

Viral Presets now run in the Higgsfield plugin for ChatGPT on GPT-6 Astra: upload a photo and get a video in a preset style, in the chat.

The plugin also covers image and video generation, branding and product graphics, Ad Multiplier, UGC and faceless video, motion graphics, and websites.

• Run effects: type @Higgsfield /effects and upload a photo.
• By name: @Higgsfield /main-character, /brand-monument, and others.
• Library: browse presets or apply one to your reference.

To get started: open the Higgsfield plugin for ChatGPT and type /effects.

Seedance 2.0 generations were failing more often than usual. The cause was a temporary provider-side issue. It's resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

GPT Image 2 generations were failing or running slow. The cause was a temporary provider-side issue. It's resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

OpenAI's flagship model now works with Higgsfield MCP: Astra handles logic and orchestration, Higgsfield generates the assets. Higgsfield Games 2.0 is built on this pairing.

• Games 2.0: multiplayer open-world games from one prompt, in any genre, 2D or 3D, through the Higgsfield plugin for ChatGPT and in Supercomputer.
• 3D: modeling, rigging, reconstruction, and interactive 3D experiences.
• Motion: motion graphics and animation.

To get started: install the Higgsfield plugin for ChatGPT, or open Supercomputer.

An AI-native 3D workspace from Higgsfield that works together with Supercomputer: a prompt or reference becomes an editable 3D scene, then video.

• Scene: objects, layout, lighting, cameras, and animation, generated from a prompt and a model of your choice, GPT-6 Astra included.
• Edits: move characters and objects, place cameras, and adjust lighting; individual objects change without regenerating the scene.
• Shots: organize shots and generate finished video in one shared workspace, with the AI agent.

To get started: open 3D Jutsu.

OpenAI's new flagship model is now in Supercomputer on Higgsfield.

Per OpenAI:

• Workflows: holds multi-step work and produces finished documents, spreadsheets, and presentations.
• Prompts: follows them closely, with clearly structured slides.
• Context: pulls only relevant context into the output, without repetition.
• Visual judgment: stronger for sites, games, apps, and renders.

To get started: open Supercomputer and select GPT-6 Astra.

Anthropic's successor to Claude Fable 5 is now in Supercomputer on Higgsfield.

Compared with Claude Fable 5, per Anthropic:

• Agentic tasks: stronger on coding, multi-step research, and document work.
• Writing: clearer, more coherent, and more accurate at following guidelines.
• Thinking: adaptive thinking is always on; reasoning depth is set with the effort selector.

To get started: open Supercomputer and select Claude Fable 5.1.

A new Higgsfield video tool that changes an object, location, product, or character in a video while leaving everything else as it was. No prompt is needed: pick from 30+ presets or briefly describe the change.

• Motion Transfer: the motion from any video, with a new scene built around it.
• Object Swap: changes what you describe; the rest of the frame is designed to stay untouched.
• Inputs: one video up to 30 seconds and up to 30 reference photos.
• Output: up to 1080p.

To get started: Video → Higgsfield Genjutsu.

Seedance 2.5 generations were failing more often than usual. The cause was a temporary provider-side issue. It's resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

MiniMax's post-trained variant of H3 is now on Higgsfield, tuned for prompt adherence and aesthetics. It caps at 768p, where the base H3 reaches 2K.

• Text-to-video: video from a description of setting, subject, action, and camera.
• Image-to-video: animates a source image, keeping its composition, identity, and aspect ratio, per MiniMax.
• Frame transitions: a start and an end frame to control the scene.

To get started: Video → MiniMax H3 Max.

Higgsfield is now available in Korean and Japanese, with no third-party translation needed.

Signed-in users can switch languages from the profile menu or Supercomputer; signed-out users can use the language selector in the footer.

Google's multimodal model for video generation and editing is now on Higgsfield, for transitions between two set frames and low-cost iteration before a final render.

• First and last frame: set both and the model generates everything in between.
• Draft mode: test ideas at 360p, then upscale the takes you keep.
• Output: up to 10 seconds per generation, at 360p, 720p, 1080p, or 4K.

To get started: Video → Gemini Omni 1.1 Flash.

Black Forest Labs' video upscaler is now in Upscale on Higgsfield. It takes any input video and outputs 1080p, 2K, or 4K, with motion and audio sync preserved, per Black Forest Labs.

• Precise: raises resolution without adding new detail.
• Creative: adds and reworks detail, with an optional prompt to steer what changes.

To get started: Video → Upscale → FLUX 3 Video Upscaler → upload a video.

Two image models from Recraft are now on Higgsfield, each for a different job.

• Recraft Styles: style transfer by reference; attach up to 10 references and following generations match that style, with no training.
• Recraft Utility: flat light, frontal composition, and simple scenes, for mockups and product shots.

To get started: Image → select Recraft V4 Styles or Recraft V4.1 Utility.

Zhipu AI’s reasoning model for code and long agentic tasks is now in Supercomputer on Higgsfield.

It first appeared in the selector under the stealth name Ox Alpha, free during that period; it now runs under its real name and uses credits.

Per Zhipu AI:

• Context: about 1M tokens, for large amounts of material in one session.
• Input: text, image, and video; output is text.
• Tools: tool calling and JSON-formatted output, for automation.
• Best for: code, long agentic tasks, and general tasks where quick responses matter.

To get started: open Supercomputer and select GLM-5.3 Flash.

A Higgsfield skill for Claude that turns one video ad into many variants: the same video, edit, and pacing, with new characters, clothing, locations, or objects. It runs through Higgsfield MCP.

• Swaps: one element or all of them per variant; the cut and pacing are designed to stay as they were.
• Audio: the original audio is kept on every non-translated variant.
• Captions: removed, or regenerated in another language with a synthetic voiceover on request.

To get started: connect Higgsfield MCP in Claude and ask for ad variants.

Share your link and earn commission on every plan bought through it.

Terms as of August 24, 2026.

• Commission: up to 25% on qualifying subscriptions, paid to your card.
• Recurring: for up to 12 months on qualifying renewals.
• Referral discount: anyone who signs up through your link gets up to 50% off for three hours.
• Tracking: see which subscriptions and commissions are attributed to your link.

To get started: apply on the Affiliate page to get your link.

The newest Wan video models are now on Higgsfield, generating from text and images, for social content, product and ad creatives, and film previz.

Per Wan:

• Output: up to 30 seconds per generation, at 480p, 720p, or 1080p.
• Consistency: faces, clothing, and details hold across frames, even with several characters.
• Audio: synced sound, with improved physics and facial expression.
• Prime: generates faster than standard 3.0, at a higher cost.

To get started: Video → select Wan 3.0 or Wan 3.0 Prime.

A shared workspace in Supercomputer that keeps context across your work, for teams and for long-running jobs.

• Teamwork: share a project; instructions are shared, but each person's chats and memory stay private.
• Skills and connectors: added to a project, they take priority over all others.
• Memory: project memory is kept separate from personal memory.
• Assets and Elements: documents live in the project; uploaded images and videos become named Elements.

To get started: open Projects in Supercomputer and create a project.

xAI's Grok Bot now works with Higgsfield MCP. Unlike a chatbot, it is built on always-on agents that carry out multi-step tasks on their own, including image and video generation on Higgsfield.

Per xAI:

• Computer: each bot runs on its own cloud computer with a browser, file system, and terminal.
• Tasks: it logs into your tools and completes multi-step tasks end to end, even when your device is off.

To get started: open Higgsfield MCP → install Grok Bot → add the Higgsfield plugin and sign in.

Generations on Kling 3.0, Kling 3.0 Turbo, Kling 3.0 Motion Control, and Kling Edit were failing. The cause was a temporary provider-side issue. It's resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

Generation and scene work are now available inside Blender through the Higgsfield plugin.

• Tools: Scene Builder, 3D Model, Character Animation, Image and Video, and Camera.
• Install: drag the .zip onto any open Blender window; it installs and enables itself.

To get started: download the .zip from the plugin page → drag it into Blender → open the Higgsfield panel and log in.

Relight and Color Palette, the video-editing tools from Cinema Studio 4.0, now also work on their own.

• Relight presets: Silhouette, Practicals, Window, Overhead Fall, Contre-jour, Soft Cross, and Auto.
• Relight custom: color, brightness, diffusion, and angle, with up to two custom light sources.
• Color Palette: 50+ grading presets, including Film Colors, Black Gloss, Candy Pink, Lime Jam, Nostalgic Blue, and Crimson Vigil, plus custom grading.

To get started: open Relight or Color Palette.

Faceless Studio is now a standalone product, previously a Supercomputer skill. It makes animated faceless videos in three steps, across four themes: Education, History, Kids, and Storytelling.

• Script: pick a topic or write your own; set format, length from 1 to 5 minutes, and animation style.
• Style: 8 presets, from Stickman Cartoon to Studio 3D.
• Voice: pick a narrator; the output is a finished episode with script, voiceover, and animation.
• Channels: style and voice are saved in your channel, so new episodes look and sound the same.

To get started: open Faceless Studio and create a channel.

Large teams get secure, dedicated access to Seedance models on Higgsfield.

• Deployment: Seedance models on dedicated US GPU infrastructure, with dedicated capacity and concurrency.
• Security: SOC 2 and SSO.
• Support: a dedicated Account Manager and a Forward Deployed Creative to onboard marketing teams.
• Team controls: unlimited seats and per-team credit allocation.

To get started: see Higgsfield for Enterprise.

Create in Public is a way for the community to back each other, and for the Higgsfield team to back creators with grants. You publish an unfinished version of your project to a public page, where people can watch it, like it, and follow its progress.

Publishing gets you:

• Shortlist of up to top 50 projects. Community support gives a project a chance to stand out before judging begins.
• Higgsfield Picks. Separate from the community top 50; the team watches published projects and can select strong work at any point in August, regardless of rating or likes.
• Credit Grants. Published projects are candidates for Higgsfield grants, up to 100k credits per project.

To get started: head to the Create in Public page and hit Create a public project.

Nano Banana Pro generations were failing more often than usual. The cause was a temporary provider-side issue. It’s resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

Version 4.0 extends duration, references, camera, color, and character performance.

• Duration: 30-second generations, enough for a full character arc.
• References: up to 50 per generation to lock a face, product, or style.
• Camera: 30+ movement presets, 4 camera types, and a rebuilt 5-lens system.
• Color Palette: 50+ presets for tonal direction without manual grading.
• Emotion Wheel: a character's performance directed with 8+ emotions.
• Era selector: pick a decade; grain, color, and lens style adjust to match.
• Tempo: control over cut speed and frequency.
• Extend: continue an existing clip forward or lead into it backward.

To get started: open Cinema Studio and start a new project.

Layers turns a flat image into separate elements you can adjust and rebuild individually.

• Layer Decomposition: splits the image into independently editable elements.
• Edit Text: finds text in an image and regenerates it with your replacements.
• Remove Background: removes the background; the result drops into your project for further editing.

To get started: open Layers and upload an image, or select an existing asset → Edit in the menu on the right.

Higgsfield is now available in Spanish for all users, with no third-party translation needed.

Signed-in users can switch languages from the profile menu, Personal Profile, or Supercomputer; signed-out users can use the language selector in the footer.

Ended: submissions closed on September 14, 2026.

The festival is live and accepting entries from any registered Higgsfield user, with no paid plan required.

• Where to begin: Create festival project on the festival page, or the Film Festival Contest section in Cinema Studio; both open a dedicated Cinema Studio project.
• Submission: generate your shots in that project, then submit the finished film with all its assets from there.

To get started: open the festival page or Cinema Studio.

Audio has a new design, a new model, Qwen, and advanced settings for emotion, format, pitch, and more.

• Text to Speech: speech from text with a preset voice, or upload or record your own.
• Voice Change: a new voice in an uploaded video, with the original delivery kept.
• Translate: translation and lip-sync into another language, from a library of 18 languages.

To get started: open Audio and pick Text to Speech, Voice Change, or Translate.

Academy was redesigned around interactive video lessons. The old courses are retired.

• Format: each course plays like a YouTube guide, split into short sections.
• Prompts: they pop up in the corner during the video; copy one on the spot or find it later in the Keypoints panel.
• Overview: a course summary and a per-lesson breakdown sit under the player.

To get started: open Academy and pick a course.

Thirteen new presets are in the Viral Presets section, built around current short-form trends.

New additions include Agamemnon, Bullet Time, Fallen Angel, Dolphin Ride, and Monet Muse.

To get started: open Viral Presets → pick a preset → upload an image → Generate.

ByteDance's new version in the Seedance video family is now on Higgsfield.

• Clips: up to 30 seconds, with ambience, foley, and score generated in the same pass as the picture.
• Region-level edits: an object or a face fixed in one spot without re-rendering the whole clip; a single shot in a sequence re-rolled separately.
• References: up to 50 per generation to lock characters, products, and style.
• Resolution: 480p and 720p, with upscaling up to 4K.

To get started: Video → Seedance 2.5.

Updated: the submission deadline later moved to September 14, 2026.

A short-film contest with a $1,000,000 prize pool. Films must be at least 3 minutes long and generated entirely on Higgsfield within the generation window.

• Eligibility: worldwide, 18+, solo or teams of up to 4.
• Prizes: 1st $500,000, 2nd $200,000, 3rd $100,000, Audience Choice $100,000, and 10 honorable mentions of $10,000 each.
• Dates: opens August 10, submissions close September 3, winners announced in the first week of October 2026.

To get started: open the festival page.

Generations on all Kling models, including Kling 3.0, Kling Avatar 2.0, Kling 2.6, and Kling Motion Control, were failing. The cause was a temporary provider-side issue. It's resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

Supercomputer now has an Auto Mode: a free model for text generation with no credits deducted. Image and video generation still spend credits.

The mode structure changed with it: Efficient was removed.

• Auto: the free text model.
• Manual: pick your own model, such as GLM, DeepSeek, Seed, or Opus.

To get started: open Supercomputer and pick Auto, or choose a model in the selector.

The Adobe After Effects plugin now works in Supercomputer and in Claude through Higgsfield MCP. The agent works directly with compositions, layers, keyframes, effects, and scripts.

• From a description: effects and animations built in your composition.
• From references: design from an image reference, animation from a video reference.
• Expressions: describe the behavior and the expression is written for you.
• Generations: Higgsfield images and video imported straight into the composition.

To get started: for Claude, install the Adobe plugin and connect it in Claude's connectors via bridge.higgsfield.ai/mcp. For Supercomputer, install the plugin; it is detected automatically.

Kling 3.0 and GPT Image 2 generations were failing instantly. The cause was a temporary provider-side issue. It's resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

Supercomputer storage can now be expanded beyond the base 20GB, paid monthly with credits: 1GB costs 2.5 credits, with the first charge at your next billing cycle.

• Pro and Plus: up to +30GB, 50GB in total.
• Max and Ultra: a slider up to +980GB (1TB in total), any value from +10GB.
• Changes: downsize or cancel from Manage Storage; changes take effect at the next billing cycle.

To get started: open Supercomputer → Manage Storage → Expand your storage, or follow the prompt when you reach the 20GB limit.

Two new models are in the Supercomputer selector on Higgsfield: GPT-5.6 from OpenAI and Grok 4.5 from xAI.

Per the providers:

• GPT-5.6: breaks complex pipelines into subtasks, holds more steps at once, and stays on track through long multi-stage runs, for work where depth matters more than speed.
• Grok 4.5: faster than earlier Grok models and cheaper at comparable quality on most technical tasks, a default for everyday pipelines.

To get started: open Supercomputer and select GPT-5.6 or Grok 4.5.

ByteDance's image model for precise work is now on Higgsfield: complex infographics, targeted edits, and photoreal textures. Also available in Supercomputer, Higgsfield MCP, and Higgsfield CLI.

Per ByteDance:

• Infographics: charts, diagrams, and photos in a single frame, with the data structure kept.
• Point-level editing: point selection, color and material swaps without rebuilding the scene.
• Group shots: several reference photos of people combined into one shot with unified lighting.
• Light: glass reflections, water splashes, and motion-blur panning.
• Text: native rendering in 10+ languages, including Arabic, Russian, Japanese, Korean, and Spanish.

To get started: Image → Seedream 5.0 Pro.

Ended: entries closed on July 22, 2026; winners were announced on July 29.

A $100,000 contest for original apps built and published in Supercomputer, any theme, from productivity tools and games to image and video generators.

• Prizes: 1st $25,000, 2nd $15,000, 3rd $10,000, and 10 honorable mentions of $5,000 each.
• Dates: entries July 8 to July 22, 2026, 8:00 AM PT; judging July 22 to 29.
• Judging: credit spend inside the app, creativity and concept, social engagement, and activity on Higgsfield.
• Entry: 18+, free with a Higgsfield account; apps had to be original, published July 8 or later, use Higgsfield model generation, and be shared on social media with #HiggsfieldApp.

Supercomputer now builds full generative apps that create images, video, and 3D through Higgsfield models; the agent wires the models in itself, with no API keys or code.

Builds spend your regular credits, up to 20 apps a day. People who use your app sign in with their Higgsfield account and generate on their own credits.

• From a description: design, code, database, sign-in, and model integration, from a landing page to a web app with logic and auth; three base templates, Studio, Stepper, and Simple App.
• Edits in chat: say "make it dark" or "add a gallery" and the agent updates the app with a fresh preview.
• Publishing: SEO and security checks, an auto-generated cover, icon, and description, like/dislike, reports, and a Credit Spend display; My Apps lists what's published.
• Also in Claude and Cursor: through Higgsfield MCP.

To get started: open Supercomputer → Products → Apps → describe your app in chat.

Storage usage is now visible in Supercomputer, with cache clearing in a couple of clicks. The default quota went up from 5GB to 20GB for all users.

To get started: open Supercomputer → Manage Account → Usage → scroll to the bottom.

The Unlimited Models Marketplace is now on Team and Scale plans, at the same prices and model selection as individual plans.

The bundle term starts at purchase, not at assignment, so unassigned time is lost.

• Per seat: bundles are bought per seat and can't be shared among team members.
• Assignment: the owner sets the seat count at checkout, then assigns bundles right away or later from Manage Account → Subscription; a bundle is assigned once and can't be reassigned.
• Roles: purchasing is owner-only; admins can assign if the owner grants the permission.

To get started: Unlimited models in the profile menu, or Manage Account → Subscription.

Availability: Team and Scale plans.

A new Higgsfield skill in Supercomputer and Higgsfield MCP for narrative video: one voiceover runs over several stylized 10-second blocks held in a single visual style.

Animated only, no photorealism: 2D, ink sketch, painterly textures, and more.

• Use: explainers and storytelling in any genre or tone, with no host on screen.
• Settings: topic or story, style, language, format, and length.
• Output: up to 10 minutes, 16:9 or 9:16.

To get started: ask Supercomputer for a narrative video; it picks up the skill automatically.

Anthropic reopened access to Claude Fable 5. It is back in Supercomputer and Higgsfield MCP, and Supercomputer Games run on it again.

• Autonomy: breaks a task into steps and carries it through.
• Pipelines: full content pipelines in Supercomputer, from idea to finished assets in one pass.
• Motion: motion design and ad videos in a consistent style across frames.
• Games: more thought-out builds in Supercomputer Games.

To get started: open Supercomputer and select Claude Fable 5.

Anthropic's new Sonnet model is now in Higgsfield MCP and Supercomputer, with autonomy that used to require heavier, pricier models.

Per Anthropic:

• Autonomy: runs full marketing-asset pipelines through Higgsfield MCP without step-by-step guidance.
• Speed: faster agentic workflows inside Supercomputer.
• Context: better context handling for structured content.
• Cost: handles tasks that used to need Opus, at a lower price.

To get started: open Supercomputer and select Claude Sonnet 5, or connect Higgsfield MCP in Claude.

Google's fast, low-cost image model at 1K is now on Higgsfield, for rapid iterations, bulk content, and social previews before a hi-res final.

Text rendering and general quality are on par with Nano Banana 2, per Google.

• References: up to 14 images, to pull a person, a scene, and a product into one composition.
• Aspect ratio: taken from your reference automatically, or picked from the full ratio set.
• Resolution: 1K only; for 2K or 4K use Nano Banana 2.

To get started: Image → Nano Banana 2 Lite.

Google DeepMind's video model is now on Higgsfield in preview (beta): 720p, clips of 4, 6, 8, or 10 seconds, 16:9 or 9:16, with native audio. It reads text, images, and video together and composes them into one shot.

Per Google DeepMind:

• Consistent characters: face, outfit, and voice hold across edits, with no recasting.
• Physics: gravity, weight, and collisions behave as in filmed footage.
• Edit by prompt: change one element of an existing clip in plain language instead of regenerating the whole scene.

To get started: Video → Gemini Omni Flash → add a prompt, an image, start and end frames, or a clip to edit.

Turn a prompt into a deployed website with a shareable link in Supercomputer on Higgsfield, with no codebase to touch.

Each build and deploy uses credits to generate assets, so heavier sites cost more.

• Full stack: front end, back end, database, auth, file storage, and API integrations.
• Assets: images and video are generated with Higgsfield inside the site, with no stock or placeholders.
• Iteration: edit meta tags and link previews, store secret keys, and remix other published sites.

To get started: open Supercomputer → describe the site, for example "build a landing page for a fitness studio". The live link is a higgsfield.app URL.

Availability: paid plans.

Buy unlimited access to a top model for a fixed window instead of spending credits per generation, so a heavy production day costs the same as a quiet one.

Generations through MCP, CLI, Canvas, and Supercomputer still use credits.

• Models: Nano Banana Pro, Seedance 2.0, GPT Image 2, Kling 3.0, and more.
• Window: 1, 7, or 14 days.
• Quality tier: where a model has tiers, the top tier includes every lower resolution.

To get started: Unlimited models in your profile menu, or Manage Account → Subscription → Get Unlimited.

Availability: Pro and Plus plans and above. Current options are on Pricing.

ByteDance's audio model is now in Audio on Higgsfield. It generates speech, sound effects, and music from a single prompt, in one pass instead of three tools and a manual mix.

• Voices: several speakers in one scene, with emotion and accent control; voice cloning takes up to 3 audio references.
• Ambience and music: rain, traffic, room tone, a musical bed, footsteps, and textures.
• References: text, an image, or an audio reference guides the result.
• Output: WAV, MP3, PCM, or OGG; length is set in the prompt, not in a duration field.

To get started: open Audio → Seed Audio 1.0 → describe the scene, including how long each element should run.

Generate a voiceover, create a custom voice, change a voice, or translate from a Claude chat, with Higgsfield Audio now in Higgsfield MCP.

To get started: connect Higgsfield MCP in Claude and ask for a voiceover, a voice change, or a translation.

Seedream 5.0 and Seedream 4.5 generations were failing more often than usual. The cause was a temporary provider-side issue. It's resolved, and generations run normally again.

Credits for failed generations were refunded automatically, so you can run them again at no extra cost.

A lighter model in the Seedance 2.0 family is now on Higgsfield, exclusive to Higgsfield with BytePlus. It runs about twice as fast as Enhanced Seedance 2.0 Fast at comparable quality, for short-form, ad cutdowns, and quick iterations.

• Resolution: 480p and 720p.
• Length: 4 to 15 seconds.
• Inputs: text, image, video, or audio references.
• Unlimited: part of the Seedance 2.0 Unlimited family alongside Enhanced Seedance 2.0 Fast.

To get started: Video → Seedance 2.0 Mini.

A 4K resolution option is now in Seedance 2.0, which removes the separate upscale step for large-screen output.

To get started: Video → Seedance 2.0 → resolution → 4K.

A bitrate setting, Standard or High, is now in Seedance models. High gives higher video quality at the same credit cost; the final file is larger.

To get started: Video → select Seedance 2.0 or Seedance 2.5 → set bitrate to High before generating.

AI Employees in Supercomputer now carry context from one session to the next, with memory and connectors.

Image, Video, and Text Generator are now Skills, called with /.

• Memory: each Employee remembers your preferences across sessions.
• Connectors: Slack, Drive, Notion, Gmail, Figma, and 30 or more integrations, for reading your docs and delivering to a folder or channel.
• Marketplace: publish an Employee, install others in one click, or import skills and memory from Claude, Claude Code, Codex, and ChatGPT.

To get started: open Supercomputer → open an Employee card to edit its instructions, files, and memory, or create your own.

Kling's variant of Kling 3.0 tuned for speed is now on Higgsfield, for locking a shot on drafts before rendering the final on standard Kling 3.0.

It supports text-to-video and image-to-video.

To get started: Video → Kling 3.0 Turbo.

Availability: on the web, in Supercomputer, Higgsfield MCP, and Higgsfield CLI.

ByteDance's upscale engine is now in Upscale on Higgsfield. It upscales video with frame interpolation up to 60fps in one pass and is tuned for both AI-generated and real footage.

• Presets: scene-aware options for AI generations, short series, UGC, and old-film restoration.
• Resolution: 2K or 4K, with no per-model tuning.
• Topaz: for face enhancement or slow motion, use the Topaz engine instead.

To get started: open Upscale → ByteDance → set resolution and preset.

Run Higgsfield tools inside Photoshop, with credits from your plan balance.

• Generation: AI Image, Realtime AI, Mockup Studio, Cinema Studio, and Marketing Studio.
• Editing: Layer Decompose, Angles, Shots, Character Swap, Remove Background, Face Swap, Upscale, Skin Enhancer, and AI Stylist.
• Requirements: Photoshop 2024 and above.

To get started: install the Photoshop plugin → Window → Extensions → Higgsfield.

Build a browser-playable game from a description on Higgsfield, with design, sound, music, characters, and multiplayer generated for you.

Every asset is generated with Higgsfield, so a game uses more credits than a single generation; a cost estimate is shown before you build.

• Where to build: in Supercomputer, or in Claude through Higgsfield MCP with the Game Studio skill.
• Publish and remix: ship to the Higgsfield Games marketplace, where others can play and remix your game.

To get started: open Supercomputer Games → pick a genre and style → describe the game. In Claude, add the Higgsfield connector and the Game Studio skill.

Run Higgsfield generation and editing tools from the DaVinci Resolve timeline, with results imported to the Media Pool or Timeline.

• Generate: video and image generation, Edit Video and Draw to Video from a prompt.
• Edit: Reframe, Remove Background, and Upscale.
• Color: Color Match builds a LUT from a reference.

To get started: install the DaVinci Resolve plugin → load a clip or reference → submit → download or import the result.

Cinema Studio now has its own workspace with its own navigation; a whole production lives in one place instead of scattered generations.

• Projects and subfolders: organize scenes and assets by shoot.
• Elements with @: pull characters and locations into the Describe your scene prompt bar.
• Sections: Home, My Generations, My Elements, My Favorites, and Community Feed.

To get started: open Cinema Studio → create or open a project → describe your scene.

Generate and edit with Higgsfield tools on the canvas in Figma and FigJam, then insert the result as a layer.

Generations use the credit balance of your Higgsfield plan.

• In Figma: image generation across models, Vector Generation (SVG), Mockup Studio, Remove Background, Color Grading, Expand, Relight, Angles, and Face Swap.
• In FigJam: Higgsfield Canvas Nodes, automatic pipelines from nodes, and chain workflows.
• Canvas input: select objects on the canvas and the plugin picks them up; auto-insert places results as they finish.

To get started: open the Plugins page → install the Figma plugin.

xAI's video model is now on Higgsfield. It animates a still image, using it as the first frame, and is designed to keep its composition and style. It also works from text alone.

• Inputs: a text prompt, with an optional start image.
• Length: 1 to 15 seconds.
• Aspect ratios: 16:9, 9:16, and 1:1.

To get started: Video → Grok Video 1.5.

Recraft's model for design work is now on Higgsfield: typography, clean lines, brand visuals, and editable SVG vectors straight from a prompt.

• Modes: Standard for general work, Vector for logos and icons, Utility for product shots, Utility Vector for brand assets.
• Vector output: SVG that scales without quality loss, in the two Vector modes.
• Brand controls: a palette of up to 10 colors, a background color, and 1K or 2K output.

To get started: Image → Recraft v4.1 → switch the Vector mode on.

A video is adapted to 9:16, 16:9, 4:3, 3:4, 21:9, or 1:1 without cropping: the model locks onto the subjects and regenerates the background instead of stretching, blurring, or cutting it.

Multi-person scenes are supported.

• References: up to 4 images for style consistency.
• Output: a video in the chosen format.

To get started: Video → Higgsfield Reframe → upload a clip and choose an aspect ratio.

Claude can now take a reference video, analyze it, and generate assets based on that breakdown in the same chat.

The Higgsfield Supercomputer skill runs inside Claude, so there’s no switching between apps needed. Generations use credits from your plan, as with all MCP requests.

• Input: any reference video dropped into the chat.
• Analysis: structure, shots, and pacing.

To get started: connect Higgsfield MCP in Claude → drop a video → ask for a recreation.

One installer adds a Higgsfield panel with five tools to both apps, so clips are processed without exporting or switching windows.

Generation runs on Higgsfield servers and uses the credits of your plan; no separate subscription is needed.

• Reframe: crop a clip to any aspect ratio with subject tracking.
• Remove Background: a clean key from any footage, no green screen.
• Upscale: up to 4K or 8K with detail restoration.
• Draw to Edit: draw over the clip to mark what to change.
• Edit Video: prompt-based editing in plain language.

To get started: install the Adobe plugin → open Premiere Pro or After Effects.

Availability: Adobe 2024 or newer.

Your own accounts on the major platforms can be linked to Supercomputer, , and the agent posts generated content to them from the chat.

Connection goes through OAuth in each network’s own login; no separate scheduling tool is needed.

• Networks: Instagram, Threads, LinkedIn, X.
• Flow: the agent generates the assets and posts them on request.

To get started: open Supercomputer → Connectors → choose a network → authorize.

Any public MCP server can now be connected by its URL directly in the chat, in addition to the built-in connector catalog.

External servers are run by their own providers; Higgsfield covers only the connection on its side.

• Connect: paste the server URL in the chat with the agent.
• Authorize: log in to the external service through OAuth.
• Manage: disconnect the same way, from the chat.

To get started: open Supercomputer → ask the agent to connect an MCP server by URL.

The same MCP server that works in Claude and Manus now connects to Cursor, so assets for a site are generated while it’s being built.

• Assets: page banners, product photos, demo and walkthrough videos, data visualizations, and motion design.
• Pipeline: build in Cursor, generate in Higgsfield, publish through Manus.

To get started: open Cursor → add Higgsfield MCP as an MCP server → sign in with your Higgsfield account.

The catalog has a second batch of presets: an image becomes a video in the chosen style.

To get started: Viral Presets → pick a preset → upload an image → Generate.

Content agents in Manus can now generate with Higgsfield models and publish the results.

Generations can be started from a phone.

• Agents: run inside Manus and use Higgsfield for images and video.
• Publishing: Meta Ads, Instagram, and the Creator Marketplace.

To get started: open Higgsfield on Manus and connect your account.

A page that collects trending video formats and in-house presets, each with a description and a matched style.

The cost depends on the preset and is shown before generation.

• Formats: cinematic game intros, Y2K paparazzi, Red Thread, Apex Hunter, In The Dark, Baseball Game, Neon City, Race Winner, and dozens more.
• Input: a photo or a video.
• Aspect ratios: 16:9, 1:1, or 9:16.

To get started: open Viral Presets → pick a preset → Generate.

An AI agent that takes a task from a single prompt through research, generation, planning, automation, and publishing.

Unlimited models don’t apply inside it; generations use plan credits.

• Memory: context carries between chats and is shown as a graph.
• Connectors: Google Drive, Notion, Slack, Gmail, and more.
• Skills: ready pipelines such as Cinema, UGC, Marketing, and Podcast.
• Scheduled tasks: a task can run automatically at set times.
• Parallel chats: several at once, with pause and resume.

To get started: open Supercomputer and describe the task.

Availability: paid plans.

A reference video becomes an ad in the same style with your own product and avatar in it.

The format is recreated automatically from the source.

• Input: an uploaded video or one from your generations.
• Assets: your product and avatar.
• Where: Marketing Studio sidebar; MCP in Claude, Hermes, and OpenClaw.

To get started: Marketing Studio → Ad Reference → upload a video and add your Product and Avatar.

Image and video generation from the terminal and from AI coding agents.

Skills run as commands or plain requests in the chat of a connected agent.

• CLI: one-line install, then higgsfield auth login.
• Skills: npx skills add higgsfield-ai/skills adds generate, soul, and more; call them with /higgsfield:generate or a plain request.
• Marketing Skills: product shot, lifestyle scene, Pinterest pin, hero banner, social carousel, ad creative pack, and more.
• Agents: Claude Code, Cursor, Codex, OpenClaw, and 12+ others.

To get started: open the CLI page and follow the install steps for your agent.

An infinite board where nodes connect into one workflow, from idea to finished media.

Boards are shared with a team in real time.

• Nodes: image and video generation, LLM, prompt, upload, text, and shapes.
• Templates: Long Video, Simple Seedance 2.0, Extend Video, Image Edit, and StoryBoard With Elements.
• Editing on the board: crop, upscale, expand, remove background, and inpaint; nodes dragged from the Feed keep their prompt and settings.

To get started: Canvas → Create Canvas, or pick a template.

Image and video generation inside the AI agent you already use, with results delivered in the same chat.

An active Higgsfield subscription and a subscription to the connected platform are required; generations use your plan’s credits, with no API key.

• Models: 16+ for images, up to 4K in any aspect ratio, and 17+ for video, up to 15 seconds with duration, genre, and start and end frame control.
• Model choice: the agent picks one for the task or uses the one you name.
• References: past generations and characters from your asset library.

To get started: in your agent, open Settings → Connectors → add https://mcp.higgsfield.ai/mcp → Connect → sign in with your Higgsfield account. Details are on the MCP page.

Alibaba’s video model is now on Higgsfield; at launch it ranked first on the Artificial Analysis Video Arena.

• Audio: video and sound generated in one pass.
• Motion: realistic physics of movement, per Alibaba.
• Storytelling: multi-shot generation.

To get started: Video → HappyHorse.

Both models now generate video in 4K.

To get started: Video → Kling 3.0 or Kling Omni → set resolution to 4K.

OpenAI’s image model for photorealistic output is now on Higgsfield.

Output goes up to 4K, with marketing visuals, mockups, and packaging as the main use cases.

Compared with GPT Image 1.5, per OpenAI:

• Text in images: more accurate rendering.
• Speed: about 2x faster.

To get started: Image → GPT Image 2.0.

Full-HD output is available alongside the existing resolutions.

To get started: Video → Seedance 2.0 → set resolution to 1080p.

A new version with project-level settings and a built-in AI director: genre, palette, camera, lens, and light are set once per project and apply to every shot.

Audio, camera movement presets, speed ramp, and the editor UI are removed in this version.

• Project settings: 7 genres, 9 color palettes, 10 camera movement styles, 7 lighting setups, three camera types, lenses from Clinical Sharp to Anamorphic, focal length 8 to 75mm, and three apertures.
• Elements: replaces Image mode. Characters, locations, and props are created as assets scoped to the project, reusable elsewhere, and carried into every shot.

To get started: Cinema Studio → set camera, genre, and style.

Seed Speech is ByteDance's engine for text-to-speech and speech recognition in one tool, aimed at dubbing, voiceover, subtitles, and AI assistants.

• TTS: voice with emotion and tone controlled through text prompts, 10 languages.
• ASR: speech recognition with automatic punctuation and speaker and emotion detection.

To get started: Audio → Seed Speech.

A tool that makes ad videos from a product photo or link, with an avatar and a format of your choice.

The whole workflow runs in one place, from product to ad.

• Input: a product photo or a link to the product page.
• Avatar: your own photo or one generated in the tool.
• Formats: UGC, Tutorial, Unboxing, Hyper Motion, Product Review, TV Spot, Wild Card, UGC Virtual Try On, and Pro Virtual Try On.

To get started: open Marketing Studio → add a product → pick an avatar and a format → Generate.

Three new models are available.

• Veo 3.1 Lite (Google): fast video at 720p or 1080p, up to 8 seconds, text-to-video and image-to-video with native audio.
• Grok Imagine Image (xAI): photorealistic and stylized text-to-image, style transfer, multi-image editing.
• Wan 2.7 (Alibaba): 1080p video up to 15 seconds, first and last frame control, 9-grid input, native audio, instruction-based editing, multi-reference support.

To get started: Video or Image → model selector.

ByteDance’s video model is now on Higgsfield in two versions: the full one and a lighter one at a lower credit cost.

• Seedance 2.0: the full model.
• Seedance 2.0 Fast: a lighter version with shorter generation time and a lower price per generation; the exact cost is on the Generate button.

To get started: Video → Seedance 2.0.

The new version is built for commercial and studio use.

Fights, chases, and explosions are generated with real weight and inertia; face, body, and clothing stay consistent across the video; sound effects, speech, and background music are generated with the video.

• Cinematic Reasoning Engine: up to 9 references plus a prompt; the model follows the narrative.
• Genres: General, Action, Horror, Comedy, Noir, Drama, Epic.
• Speed ramps: Linear, Auto, Flash In/Out, Slow-mo, Bullet Time, Impact, Ramp Up.
• Content: everything is created inside the studio; characters and locations from Cinema Studio 2.5 can be reused.

To get started: Cinema Studio → select Cinema Studio 3.0.

A streaming section for series and films made entirely with generative AI, by Higgsfield and independent creators.

Episodes are free to watch, and you can set reminders for upcoming releases.

• Catalog: Higgsfield Original works across genres and worlds, plus the team’s picks.
• React: like, dislike, share, and comment.
• Getting featured: through Higgsfield contests.

To get started: open Originals and pick a series.

AI actors can now be cast directly in a keyframe, and color grading is built into the studio.

The cycle from casting to post-production runs in one place.

• AI Cast: up to 3 actors in a keyframe, with realistic skin and expressions.
• Color grading: temperature, contrast, exposure, grain, and the final look.
• Workflow: pick a character and a location, generate, then grade.

To get started: Cinema Studio → add AI Cast actors → Generate.

A virtual casting studio for AI actors: an actor’s look and character are set before the shot.

• Film settings: genre, era from the 1900s to today, and budget, which affects polish and image quality.
• Archetypes: 12 classic types such as Hero, Rebel, and Sage.
• Appearance: gender, age, build, eye color, hairstyle, and more.
• Details: scars, tattoos, freckles, or custom features through manual input.
• Style: from vintage and uniforms to high fashion.

To get started: open Cinema Studio → switch to the Image mode → select AI Cast.

The page now shows credit grants, deductions, and a record for renewals, resets, plan changes, and pack purchases.

• Renewals and resets: old credits expire and new ones arrive, each on its own line.
• Plan changes: the deduction of the old plan’s balance and the grant of the new one.
• Packs: each purchase as a separate line.

To get started: open Manage Account → Usage History.

A 15-image set of the same character in different poses, outfits, and locations, generated in one run.

All frames are based on one Soul ID, with a mix of selfies, candid shots, golden-hour and city scenes that reads like a camera roll. The set is ready to post as a carousel.

• Presets: Male Archive, Cool Girl Dump, Surreal Self, and more.
• Character: an existing Soul ID, or a new one created for the set.
• Output: 15 images per generation, in a couple of minutes.

To get started: Image → Soul 2.0 → Photodump → pick a preset and a character → Generate.

Custom visual styles trained on your own reference images, plus 20+ curated ones such as Y2K and Digital Camera.

A moodboard carries the tone, palette, and mood of a generation without describing them in the prompt. It replaces Presets; the previous ones are kept as Curated Moodboards.

• Personal: at least 10 images in one consistent style, without blur and ideally without faces.
• Curated: 20+ built-in styles.
• Works with: Soul ID, Soul Reference, and Soul Hex.

To get started: Image → Soul 2.0 → Moodboards → Create.

Google’s new model in the Nano Banana line is now on Higgsfield: image generation and conversational editing at Flash speed.

Compared with Nano Banana Pro, per Google:

• Speed and cost: the same level of output quality with faster generation and a lower price.
• Text: more accurate and legible text in images, including text in other languages.
• References: up to 14 images per request for editing.
• Formats: 1K, 2K, and 4K output, plus new 1:4, 4:1, 1:8, and 8:1 aspect ratios.

To get started: Image → Nano Banana 2.

A color style extracted from your own images and applied to new Soul 2.0 generations, so a series keeps one palette without describing colors in every prompt.

It works inside Soul 2.0 only.

• Input: up to 15 images; the palette preview and cover are generated automatically.
• Signature: saved under a name, for example Blue Haze, in your library and listed in the Soul 2.0 model settings.
• Presets: Vibrant, Muted, Neon, Earthy, and others are included.

To get started: Image → Soul 2.0 → Color Transfer → Upload & Create.

A personal asset library: characters, locations, outfits, and props are saved once and called in any prompt with @.

Each element keeps its reference images under one tag; typing @ in the prompt inserts it, and the model picks up the visual details without a written description.

• Categories: Characters, Locations, and Props.
• Creating: a reference image or a video; a video element can keep, remove, or replace its audio.
• Supported models: Kling 3.0, Seedream 4.5 and 5.0 Lite, Cinema Studio image and video, Nano Banana Pro, and more.

To get started: open a supported model → Elements → Create → upload references → type @ in a prompt.

ByteDance’s newest image model in the Seedream line is now on Higgsfield.

Besides generating from a prompt, it can search the web and pull current events, memes, and news into an image.

Compared with earlier Seedream models, per ByteDance:

• Prompt understanding: picks up intent, mood, and context from short phrases.
• Consistency: a character’s face, clothing, and details hold better across images.
• Editing: background or lighting changes from a plain text instruction.

To get started: Image → Seedream 5.0 Lite.

The second version of Higgsfield’s own photorealistic image model, retrained for fashion references, subcultural codes, and internet aesthetics.

• Style: picks up a look from a reference image, from streetwear and runway to online aesthetics, without a written description.
• Characters: Soul 1 characters are upgraded separately.

To get started: Image → Soul 2.0.

A page with all active challenges and contests: brief, rules, deadline, prize, and winners’ entries in one place.

Each contest runs through multiple stages, with winners picked by a jury; several contests can run at once.

• Inside a contest: theme and details, a submission guide, the deadline, the prize, and an FAQ.
• Entries: view and like other participants’ work.

To get started: Contests → open an active challenge or a contest → Submit.

A section with credit spend, usage per model, and a log of every charged generation.

• Spend Overview: total credits spent and number of generations.
• Credits Usage: spend split by model.
• History: each charge, per generation.

To get started: Manage Account → Usage.

Kling’s new generation is now on Higgsfield: a video model, a multimodal Omni variant with native audio, and targeted editing of existing clips.

Per Kling:

• Kling 3.0: video from a prompt or an image, clips from 3 to 15 seconds at 720p or 1080p.
• Kling Omni: text, images, video, and audio in one model; generates dialogue, sound effects, and lip-synced speech with the video, keeps characters consistent from references, and handles multi-character scenes.
• Kling Omni Edit: changes to a specific part of an existing clip without regenerating the whole video.

To get started: Video → model selector → Kling 3.0, Kling Omni, or Kling Omni Edit.

Image and video generations can be canceled while they wait in the queue.

If Cancel returns “Failed to cancel generation,” the generation has already started and will complete normally.

• How: hover over a queued generation → Cancel.
• Scope: works only while the job is still in the queue.

xAI’s video model is now on Higgsfield, aimed at short clips with a story, voice, and motion.

Per xAI:

• Storytelling: short narrative videos.
• Brand content: marketing and brand clips.
• Voice: scenes with voiceover and emotional delivery.
• Motion: realistic movement and camera work.

To get started: Video → Grok Imagine.

The camera around a subject can now be set directly by dragging, instead of fixed options.

• Drag to rotate: hold and drag to change the camera direction.
• Orbit: a visual control that shows the camera position.
• Quick angles: Top, Front, Left, Right, and more.

To get started: Image → Angles.

Ultra HD is now available as a quality option for Google's Veo 3.1. Pick 4K in the quality settings before generating.

To get started: Video → Veo 3.1.