ChangelogThe latest
from Higgsfield

Seedream 5.0 Flash is a new model in Seedream image line: fast image generation with intelligent visual reasoning.

• Resolution: 1K, 1.5K, and 2K.
• Multi-reference: add several references and the model blends their style in one frame.
• Generation and editing: create from scratch or fix an existing image without switching models.

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

Claude Opus 5.5, the new version of Anthropic's Opus model, is now in Supercomputer.

• Stronger than Opus 5 in agentic coding, long multi-step tasks, and knowledge work: reports, analysis, financial models.
• More efficient: it completes typical tasks using fewer tokens.
• Clearer output: reports on what it did, found, and needs next in plain language, with less filler.
• Better visual reading: charts, diagrams, and screenshots parse more accurately, and computer use at default effort matches Opus 5 on high.

To get started: open Supercomputer and select Claude Opus 5.5.

Seedance 2.5 Draft Mode is a new model in Seedance video generation for drafting before the final. Generate previews at 480p, pick the take that works, then generate that one at 1080p on Seedance 2.5.

• 480p drafts: check composition, motion, pacing, and audio-visual timing; adjust the prompt and references before the final.
• 1080p final: run the selected draft's prompt, references, seed, aspect ratio, and duration on Seedance 2.5.

Availability: all regions except the US, where the model is unavailable on the provider's side.

To get started: Video → model selector → Seedance 2.5 Draft Mode.

GPT Image 2.5, OpenAI's new image model, is live in two variants.

• GPT Image 2.5 Flare: the default for most tasks. Higher quality than GPT Image 2 at up to 50% lower latency.
• GPT Image 2.5 Sunburst: for precise control in edits; generation takes longer.

Across both: sharper detail, better subject preservation from references, targeted edits that don't degrade across iterations, and stronger style holding.

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

OpenAI has expanded the GPT-6 line with two models, trained with the same methods as GPT-6 Astra but tuned for price and speed. Both are now available in Supercomputer and Higgsfield MCP.

• 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.
• GPT-6 Astra stays the flagship, for when you need the strongest result.

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

Grok 4.7, SpaceXAI's most capable model for coding and knowledge work, is now available in Supercomputer.

• Larger than Grok 4.6 and trained longer on multi-hour tasks: it checks its own output more carefully and holds longer context.
• Tuned for conversation as well as code, with fewer slips on everyday work tasks.
• Better at documents and presentations.

To get started: open Supercomputer and select Grok 4.7.

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

• Pay per generation. No subscriptions or seats; public per-model rates, only successful generations are billed. Separate from higgsfield.ai plans and credits.
• Three calls: submit a prompt, poll the status, fetch the file. SDKs for Python and TypeScript, or plain REST.
• Launch offer: 15% off all models, up to 50% off on 4 models of your choice.

To get started: create a key at Higgsfield API.

The presets run on GPT Astra paired with Higgsfield through the ChatGPT plugin: seven motion workflows, each with a ready prompt, a workflow breakdown, and a Try in ChatGPT button. Also available as the Motion Design Bundle in one set.

Presets: Illustration Animation, Localization Motion, Motion Design, Paper Collage, Presentation Animation, SaaS Animation, Whiteboard Animation.

To get started: select a preset from the Presets Library and hit Try in ChatGPT.

AI Motion Designer is a new agent that connects Higgsfield to Adobe After Effects through GPT. Generate visuals in Higgsfield, describe the motion work in text, and the agent builds it in your open AE composition: titles, logo reveals, transitions, promos, effects.

• Works from your material and references; translates the request into layers, keyframes, and timing.
• Resolution (1080p to 8K) and duration are set in the prompt.
• Output is native AE elements, editable as usual.

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

Requirements: active Higgsfield, GPT, and After Effects subscriptions; generations use your regular credits.

Effects 2.0 is live in the ChatGPT plugin, running on GPT-6 Astra: upload a photo and get a video in one of the viral preset styles, in the chat.

• Run effects: type @Higgsfield /effects and upload a photo.
• Call a preset by name: @Higgsfield /main-character, /brand-monument, and others.
• Browse the library or apply a preset to your reference.

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

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

GPT-6 Astra, OpenAI's new flagship model, is now available through Higgsfield MCP: use Astra and Higgsfield together, with Astra handling logic and orchestration and Higgsfield generating the assets. Higgsfield Games 2.0 is built on this pairing.

Higgsfield Games 2.0

• Build and deploy multiplayer open-world games from one prompt, in any genre, 2D or 3D.
• Works through ChatGPT (Higgsfield Plugin) and in Supercomputer.

Also with MCP and Astra

• 3D modeling, rigging, and reconstruction.
• Motion graphics and animation.
• Interactive 3D experiences.

To get started: install the ChatGPT Plugin, or use Supercomputer directly.

3D Jutsu is an AI-native 3D workspace from Higgsfield that works together with Supercomputer.

• Turns a prompt or reference into an editable 3D scene, with objects, layout, lighting, cameras, and animation, and then into video.
• Edits individual objects in the scene without regenerating everything.
• Organize shots and generate finished video in one shared workspace.

How it works

1. Describe what you want in a prompt and pick a model, GPT-6 Astra included.
2. Get a draft scene: move characters and objects, place cameras, adjust lighting.
3. Build it all together with the AI agent.

To get started: open 3D Jutsu.

GPT-6 Astra, OpenAI's new flagship model, is now available in Supercomputer.

• Holds multi-step workflows and produces finished documents, spreadsheets, and presentations.
• Follows prompts closely, with clearly structured slides.
• Pulls only relevant context into the output, without repetition, so results are usable as is.
• Stronger visual judgment for sites, games, apps, and renders.

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

Scale, our Business plan for teams of 5 to 15, now includes unlimited models. Existing subscribers keep their original terms; the changes apply to new subscriptions.

Unlimited models on Scale

• Monthly plan: Nano Banana Pro, Seedream 5.0 Pro.
• Annual plan: Nano Banana Pro, Seedream 5.0 Pro, and Kling 3.0.
• Unlimited access is granted for 7 days after the first purchase.

To get started: see Pricing.

Claude Fable 5.1, a new Anthropic model and the successor to Claude Fable 5, is now available in Supercomputer.

• Stronger on long agentic tasks: coding, multi-step research, and document work.
• Improved writing: clearer, more coherent, and more accurate at following guidelines.
• Adaptive thinking is always on; set reasoning depth with the effort selector.

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

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

• Motion Transfer: take the motion from any video and build a new scene around it.
• Object Swap: change exactly what you describe; the rest of the frame stays untouched.

Inputs: one video (up to 30 seconds) and up to 30 reference photos. Output up to 1080p.

To get started: Video → model selector → choose Genjutsu.

MiniMax H3 Max is live, a post-trained variant of MiniMax H3 tuned for prompt adherence and aesthetics.

• Text-to-video: generate video from a text description of setting, subject, action, and camera.
• Image-to-video: animate a source image, keeping composition, identity, and aspect ratio.
• Directed frame transitions: set a start and end frame to control the scene.
• Built for speed over resolution: caps at 768p, where base H3 reaches 2K.

To get started: Video → model selector → choose MiniMax H3 Max.

Gemini Omni 1.1 Flash is live, Google's multimodal model for video generation and editing.

• First and last frames: set a starting and ending frame, and the model generates everything in between.
• Draft mode: test ideas at 360p, cheaper and faster, then upscale the takes you keep.
• Up to 10 seconds per generation, at 360p, 720p, 1080p, or 4K.

Best for: transitions and camera moves between two set frames, and fast, low-cost iteration before a final render.To get started: Video → model selector → choose Gemini Omni 1.1 Flash.

FLUX 3 Video Upscaler raises the resolution of video while keeping motion smooth and audio in sync. It accepts any input video and outputs at 1080p, 2K, or 4K.

Two modes

• Precise. Raises resolution without adding new detail.
• Creative. Adds and reworks detail, with an optional prompt (detail guidance) to steer what gets changed.

To get started: open Upscale and upload a video.

Higgsfield MCP calls were failing earlier on August 27. Resolved, and MCP is working normally again.

Three new Recraft image models are live: V4.1, Styles, and Utility.

• Recraft V4.1. The main model. Varies light, mood, and composition across generations.
• Recraft Styles. Style transfer by reference. Attach up to 10 references and following generations match that style. No training required.
• Recraft Utility. Flat light, frontal composition, simple scenes. For mockups and product shots.

To get started: Image → model selector → Recraft.

Apps in Supercomputer are back up and running normally.

Ox Alpha's stealth period has ended: it's now revealed as GLM-5.3 Flash and costs credits. The same model is available in Supercomputer under its real name.

• A fast, lightweight model from Zhipu AI's GLM line.
• Tuned for speed and low cost, suited to chat, code, and general tasks where quick responses matter.

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

• Create in Public deadline: now September 12
• Final submission deadline: now September 14

All following dates shift accordingly. See the festival page for the full timeline.

Ad Multiplier turns one video ad into many variants: same video, edit, and pacing, with new characters, clothing, locations, and objects. Runs through the Higgsfield MCP connector.

• Same video, new details. Swap one element or all of them per variant; performance, cut, and pacing are preserved.
• Original audio kept on every non-translated variant.
• Captions can be removed, or regenerated in another language on request (with a synthetic voiceover).

To get started: connect Higgsfield to Claude via MCP.

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

• 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.
• Referral tracking. See which subscriptions and commissions are attributed to your link.

To get started: apply at Higgsfield to get your link.

Ox Alpha, a new reasoning model, is now free in Supercomputer. Built for code, long multi-step tasks, and large context.

• Reasoning. Built for coding, sustained agentic work, and production tasks, with step-by-step logic across long jobs.
• Large context. A roughly 1M-token context window for working over large amounts of material in one session.
• Multimodal. Accepts text, image, and video input, and returns text.
• Tool calling and structured output. Calls external tools and supports JSON-formatted output, useful for automation.

To get started: open Supercomputer and select Ox Alpha in the model selector.

The newest Wan video models are live, generating video from text and images.

• Up to 30 seconds per generation, at 480p, 720p, or 1080p.
• Multi-subject consistency: faces, clothing, and details stay stable across frames, even with several characters.
• Audio sync, with improved handling of physics and facial expression.
• Wan 3.0 Prime generates faster than earlier models, including standard 3.0, at a higher cost.

Best for: social media content, product and ad creatives, and film previz.

To get started: Video → model selector → choose Wan 3.0 or Wan 3.0 Prime.

Projects is a shared workspace in Supercomputer that keeps context across your work.

• Teamwork. Share a project with others; instructions are shared, but each person's chats and memory stay private.
• Project skills and connectors. Add skills and connectors to a project, where they take priority over all others.
• Project memory. Kept separate from personal memory, so nothing personal carries in or out.
• Assets and Elements. Store documents in the project; uploaded images and videos become named Elements.

To get started: open Supercomputer and create a project.

Grok Bot by xAI is now in Higgsfield MCP. Unlike a chatbot, it's built on always-on AI agents that carry out multi-step tasks on their own.

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

To get started: head to Higgsfield MCP, install Grok Bot, then add the Higgsfield plugin and sign in. Once connected, just ask it to generate an image or video.

New users: verify a card to start the MCP 3-day trial and get 100 MCP-only credits. If not cancelled before the end of the trial, it renews as a monthly Plus Plan.

Higgsfield is now available inside Blender, the 3D editor for modeling, animation, and rendering, with generation and scene work built right into the plugin.

Tools: Scene Builder, 3D Model, Character Animation, Image and Video, and Camera.

How to install: download the .zip from the plugin page, drag it onto any open Blender window (it installs and enables itself), then open the Higgsfield panel, log in, and start generating.

Relight and Color Palette are now available as standalone video-editing products. Same functionality as in Cinema Studio 4.0, now also usable on their own.

Relight

• Six presets: Silhouette, Practicals, Window, Overhead Fall, Contre-jour, and Soft Cross, plus Auto.
• Or build your own with control over color, brightness, diffusion, and angle.
• Add up to two custom light sources.

Color Palette

• 50+ color-grading presets, including Film Colors, Black Gloss, Candy Pink, Lime Jam, Nostalgic Blue, Soft Palette, Overtime, 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 generates 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 (1–5 min), and animation style.
• Style. Choose from 8 presets, from Stickman Cartoon to Studio 3D.
• Voice. Pick a narrator; 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.

Higgsfield for Enterprise gives large teams secure, dedicated access to Seedance models.

• Dedicated US deployment. Enterprise access to Seedance models on dedicated US GPU infrastructure, with dedicated capacity and concurrency.
• Security and access. SOC 2 and SSO.
• Hands-on support. A dedicated Account Manager and a Forward Deployed Creative to onboard marketing teams into AI-native production.
• Team controls. Unlimited seats and per-team credit allocation for control over each project's unit economics.

Learn more: higgsfield.ai/enterprise

Seedance 2.5 now generates in 1080p on Higgsfield, in early access. It's included with Plus plans and above.

Limited time offers:

• Free trial for new users. Five 1080p generations, up to 20 seconds each, over 24 hours. No charge, card verification only.
• Bonus Seconds for subscribers. Generate Seedance 2.5 at 1080p at a reduced credit cost. Each pack is a pool of seconds billed by clip length, so a 10-second video spends 10 seconds from the pool, leaving credits untouched. Valid 30 days.

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 had a high failure rate from a temporary Google infrastructure issue. Resolved on the provider's side.

Cinema Studio 4.0 is live. Improved duration, references, camera, color, and character performance.

• 30-second generations. Up from before, enough for a full character arc.
• Up to 50 references per generation to lock a face, product, or style.
• Camera. 30+ movement presets, 4 camera types, and a rebuilt 5-lens system.
• Colour Palette. 50+ presets for tonal direction without manual grading.
• Emotion Wheel. Direct a character's performance with 8+ emotions.
• Era selector. Pick a decade; grain, color, and lens style adjust to match.
• Tempo control over cut speed and frequency.
• Forward / Backward Extend. Continue or lead into an existing clip.

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

Layers turns any flat image into separate, editable elements you can adjust and rebuild individually.

• Layer Decomposition. Splits a flat 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 straight into your project for further editing.

To get started: go to Layers and upload a new image, or select an existing asset and choose Edit in the menu on the right.

Higgsfield is now available in Spanish. Users can switch the site to Spanish directly, with no third-party translation needed.

• Signed-in users. The switch is available in the profile menu when hovering over the profile icon, in Personal Profile, and in Supercomputer.
• Signed-out users. The language switch is in the footer, at the bottom of the page.

The Global Film Festival is now live and accepting entries. Submissions close September 3.

• Where to begin. Two ways: click Create festival project on the festival page, or open the Film Festival Contest section in Cinema Studio and create your folder there. Both open a dedicated Cinema Studio project.
• Build and submit in one place. Generate your shots in that project, then submit the finished film with all its assets from there.
• No paid plan required. Any registered Higgsfield user can enter.

To start: open the Festival page or head to Cinema Studio.

Audio has been redesigned. A new model, Qwen, has been added, along with advanced settings for audio models: control over emotion, format, pitch, and more.

• Text to Speech. Generate speech from text using a preset voice, or upload or record your own.
• Voice Change. Swap the voice in an uploaded video while keeping the original delivery.
• Translate. Translate and lip-sync a video into another language, from a library of 18 languages.

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

Academy has been redesigned. The old courses are retired and replaced by interactive video lessons. Completion no longer comes with a certificate.

• Courses play as video. Each course runs like a YouTube guide, split into short sections so you can move at your own pace.
• In-video prompts. Prompts pop up in the corner during the video; copy one on the spot or find it later in the Keypoints panel.
• Overview under the player. A short course summary and a per-lesson breakdown of what each lesson covers sit below the video.

To get started: open Academy and pick a course.

Thirteen new presets are live 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 → get a video in that style.

Seedance 2.5 from ByteDance is available on Higgsfield, the latest version in the Seedance video family.

• 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 can be fixed in one spot without re-rendering the whole clip; a single shot in a sequence can be re-rolled separately.
• Up to 50 references per generation to lock characters, products, and style.
• Resolution. The model generates at 480p and 720p, with upscaling up to 4K.

To get started: Video → model selector → Seedance 2.5.

Availability: mid and upper plans include a 33-day unlimited window on Seedance 2.5, with clip-length limits varying by plan.

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. Open 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 the first week of October.

To get started: open the Higgsfield Global Film Festival contest page.

Get Unlimited access to all our top Image, Video, and Audio models at once, including Nano Banana Pro, Seedance 2.0, Gemini Omni Flash, GPT Image, and more.

• One pass to everything. Image, video, and audio top models covered by a single Unlimited offer.
• Three ways in. A 1-day free trial for new users, 7 days included on premium monthly plans, or a bundle for existing plans.

To get started: start the 1-day free trial, pick an eligible plan on Pricing, or buy a bundle from the Unlimited models button in your profile menu.

Availability: limits vary by plan, see the Pricing page. Unlimited runs 1 image, video, or audio at a time, on a shared concurrency across models.

Some users couldn't load the site, mainly in Chrome, while mobile and Safari worked. Fixed on the provider's side, everything is running normally again.

Supercomputer now has a Free Mode: a free LLM for text generation, with no credits deducted.

• Fully free text. All text generations on the Free model cost nothing, and there are no usage limits for now. Image and video generation still spend credits as usual.
• New mode structure. Auto and Efficient are gone: Free (free model), Smart (top models for complex reasoning), and Manual (pick your own model: GLM, DeepSeek, Seed, Opus, etc.).
• Effort control in Manual. Choose Low/Medium/High/Extra to set how long the model "thinks": lower effort is faster, higher effort means deeper reasoning and better results.

To get started: pick Free, Smart, or Manual in the Supercomputer mode selector.

The Filmmaker Grant is a 2-week training program for the most talented creators, hand-picked by the Higgsfield creative team.

• What selected creators get. 100,000 credits: the full grant lands on the account on selection day, no daily limits, plus a 2-week program with live sessions from the creative team and a private Discord with shot breakdowns and direct feedback.
• Who can apply. 18+, worldwide, any creative background. Required: a certificate for The AI Filmmaking Pipeline course in Higgsfield Academy.
• How selection works. Every application is reviewed manually by the creative team, there's no single winner. Selected creators are contacted directly.

To get started: complete The AI Filmmaking Pipeline in Academy and claim your certificate → prepare a showreel of your strongest work (any skill, any tools, with or without AI) → post it on Instagram, X, LinkedIn, or Reddit with the caption "Higgsfield Filmmaker Grant: Why I Want It" → submit the post link through the application form.

Higgsfield Academy is a new learning hub on the platform: practical workflows from working AI filmmakers, finish a course, take a real brief, join the crew. Built for total beginners and future power users alike.

• Watch & Learn. The latest YouTube how-to guides from our creative team: tips and workflows for making ultra-realistic content with our newest releases.
• Interactive courses. From tool onboarding to making your own film. Interactive and teacher-free: many lessons embed the actual product plus quizzes, and each course ends with a final exam. Two Cinema Studio courses are live: Getting Started with Cinema Studio (Beginner, 12 lessons, 26 min, 8 Free generations) and The AI Filmmaking Pipeline (Medium, 22 lessons, 58 min, 50+ Free generations).
• Free to learn. All you need is a Higgsfield account, free generations are built into every course, so your own credits aren't spent.
• Certificates. The AI Filmmaking Pipeline course grants a certificate with a unique credential ID, postable straight to LinkedIn.
• Filmmaker Grant, right on the page. The certificate for The AI Filmmaking Pipeline is the ticket in: learn the craft → cut your showreel → post it and tell us why it should be you → get selected for 100,000 credits, a private crew Discord, and live trainings.

To get started: open Academy on the platform and start with a Cinema Studio course.

Update: the referral program ended on July 16, 2026.

Invite friends and earn up to 7 days of Unlimited Seedance 2.0 (720p, 8 sec), while your friends get up to 60% off their plan purchase.

• How it works. Generate your unique referral link, a 24-hour window starts. If a friend follows it, purchases a Plus/Pro monthly plan or higher, and makes 1 credit-paid generation within that window, you get the reward.
• More friends, more days. 1 friend → 3 days, 2 friends → 5 days, 3 friends → 7 days (the maximum).
• After 24 hours. The link can't be recreated, but the block with invited friends stays visible; if all friends complete the steps, a modal confirms your reward.

To get started: click on the «Refer a friend» button in your profile menu, generate your link, and share it.

Availability: on Pro/Plus plans and above. Not available on Starter/Basic and Team plans.

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

• Effects and animations from a description, built right in your composition.
• Recreate from references. Design from an image reference, animation from a video reference.
• Expressions on demand. Describe the behavior and get the expression written for you.
• Higgsfield generations imported straight into the composition.

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

A new rewards system gives you bonuses for completing tasks and staying active on the platform: free gens, credit cashback, and Unlimited days.

• Tasks in your notifications. Find them left of your profile icon, complete them, and claim the reward.
• Real rewards. Spend 1,000 credits → free gens on Nano Banana Pro 2K. Generate 10+ minutes on Seedance 2.0 → 1 day of Unlimited Seedance 2.0 Fast 720p. Generate 15 days in a row → 5 days of Unlimited NBP 2K.
• Grand prize. Complete every task on time and monthly Pro and higher plans get 7 days of Unlimited Nano Banana Pro 2K right after renewal; annual plans get the same after 2 full months without cancelling.

To get started: open your notifications and check your active tasks.

Availability: on Pro/Plus plans and above. Claim rewards in your notifications before your next renewal, unclaimed rewards expire.

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

• Pro/Plus: up to +30GB extra, 50GB total.
• Max/Ultra: an adjustable slider up to +980GB extra (1TB total), any value from +10GB and up.
• Manage anytime. Downsize or cancel from Manage Storage, changes take effect at the next billing cycle.

To get started: hit Expand your storage in Manage Storage, or when you reach the 20GB limit.

Curated moodboards in Soul 2.0 weren't loading. Resolved, a quick refresh brings moodboards back to normal.

Two new models in the Supercomputer selector.

• GPT‑5.6. OpenAI's most powerful model to date: breaks complex pipelines into subtasks, holds more steps in mind at once, and stays on track through long multi-stage runs. Pick it for the most demanding work where depth matters more than speed.
• Grok 4.5. xAI's new model with a strong engineering base: noticeably faster, well balanced between speed and quality, and cheaper at comparable quality on most technical tasks, a solid default for everyday pipelines.

To get started: pick GPT‑5.6 or Grok 4.5 in the Supercomputer model selector.

Seedream 5.0 Pro from ByteDance is built to translate the creator's intent into the final visual with precision: complex infographics, surgical edits, and photoreal textures.

• Infographics in one pass. Charts, diagrams, and realistic photos in a single frame, with logical structure and no data loss.
• Point-level editing. It understands what sits where in the frame: point selection, color and material swaps without rebuilding the scene.
• Group shots from references. Combines several reference photos of people into one shot with unified lighting and texture.
• Real light physics. Glass reflections, water splashes, and motion-blur panning.
• Native text in 10+ languages. Including Arabic (right-to-left), Russian, Japanese, Korean, and Spanish, with accurate local typography.

To get started: Image → model selector → Seedream 5.0 Pro. Also available through Supercomputer, MCP, and CLI.

The site was down from 13:38 to 13:52 UTC. Resolved, everything is running normally again.

To celebrate the Higgsfield Apps launch, the App Contest is live with a $100,000 prize pool. Build and publish an original app in Supercomputer, any theme, from productivity tools and games to image and video generators, as long as people want to generate in it.

• Prizes. 1st: $25,000 · 2nd: $15,000 · 3rd: $10,000 · 10 honorable mentions: $5,000 each.
• Dates. Entries July 8, 8:00 AM PT – July 22, 2026, 8:00 AM PT. Judging July 22–29. Winners announced July 29.
• Judging. Credit Spend inside the app, creativity and concept, social engagement, and activity on Higgsfield.

To get started: build and publish an app in Supercomputer, then share it on social media with a link and the #HiggsfieldApp hashtag to complete your entry.

Availability: 18+, free to enter with a Higgsfield account. Apps must be original, published July 8 or later, and use Higgsfield model generation, apps running generation through third-party APIs are disqualified.

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

• Describe it, get it. The agent handles design, code, database, sign-in, and model integration, from a landing page or portfolio to a full web app with logic and auth (AI photobooth, anime filter, generative tools). Three base templates: Studio, Stepper, and Simple App.
• Edit in the same chat. Say "make it dark" or "add a gallery" and the agent updates the app with a fresh preview in minutes.
• Users pay for their own generations. People sign in with their Higgsfield account and generate on their own credits, so a thousand users cost the creator nothing. Builds spend your regular credits, up to 20 apps a day.
• Publish and remix. One-click SEO and security checks at publish, auto-generated cover, icon, and description (editable), like/dislike, reports, and Credit Spend display. The new My Apps section shows what's published, with hover editing.

To get started: Apps now live in the Products section in Supercomputer. Describe your app in chat and the agent does the rest. Also available through MCP, so you can build straight from Claude or Cursor.

You can now see your Supercomputer storage and clear the cache right from the interface, and the default storage quota went up from 5GB to 20GB for all users.

• Storage at a glance. See how much space is used.
• Cache cleared in a couple of clicks.

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

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

• Per-seat purchase. Bundles are bought per seat, a single bundle can't be shared among multiple team members.
• Assign after purchase. The owner sets seat count at checkout, then assigns bundles right away or later from Manage Account → Subscription.
• Assign once. A bundle can only be assigned one time and can't be reassigned to someone else.
• Owner and admin can assign. Purchasing is owner-only; admins can assign if the owner grants that permission.

To get started: Unlimited models button in the profile menu, or Manage Account → Subscription; same entry points as on individual plans.

Availability: the bundle term starts at purchase, not at assignment, so unassigned time is lost if not assigned before it expires.

Explainer is no longer just a Supercomputer/MCP skill, it's a standalone tool on the platform, with a full visual interface.

• Full creative control. Set style, character, voice, and motion yourself, no agent needed.
• Preset library built in. Pick a visual style right in the interface.
• Everything in one screen. Topic, narrator voice, format, and length, then generate in one click.

To get started: open Higgsfield Explainer → pick a preset → describe your topic (optionally add a photo or link) → set aspect ratio, duration, and voice → Generate.

Availability: up to 10 minutes, 16:9 or 9:16.

A new skill in Supercomputer and MCP for narrative video: one voiceover runs over several stylized 10-second blocks, held together in a single visual style; animated only, no photorealism (2D, ink sketch, painterly textures, and more).

• Built for explainers and storytelling in any genre or tone.
• Faceless-friendly. Visual and voiceover carry the video, no host required.

To get started: ask Supercomputer for a narrative video, it picks up the skill automatically. Describe your topic or story, choose style, language, format, and length.

Availability: up to 10 minutes, 16:9 or 9:16.

The site went down twice, from 5:45–5:48 and 5:55–6:16. Resolved, everything is running normally again.

Anthropic reopened access to Claude Fable 5! It's back in Supercomputer and MCP, and Supercomputer Games are powered by Fable 5 once again.

• Plans and finishes on its own. Breaks a task into steps and carries it through without hand-holding.
• Full content pipelines in Supercomputer, from idea to finished assets in one pass.
• On-brand motion. Motion design and ad videos that hold a consistent style across frames.
• Better games. Higher-quality, more thought-out builds in Supercomputer Games.

Claude Sonnet 5 is live in Higgsfield MCP and Supercomputer: a new level of autonomy that used to require heavier, pricier models.

• Plans and executes on its own. Runs full marketing-asset pipelines through Higgsfield MCP without step-by-step guidance.
• Faster agent pipelines. Speeds up and streamlines agentic workflows inside Supercomputer.
• Sharper output. Better context handling for more precise ideas and structured content.
• Opus-level results, lower cost. Handles tasks that used to need Opus, at a fraction of the price.

To get started: pick Claude Sonnet 5 in the model selector in Supercomputer or MCP.

Shorts Studio restyles your own footage into a completely new look, powered by Gemini Omni Flash. Upload a clip, pick a style, and get a finished short on the other end.

• 44 ready-made presets, or build your own from a photo/video moodboard and Shorts Studio picks up the style.
• Motion stays yours. Original pacing and performance carry over, only the look changes.

To get started: Video → Shorts Studio → upload a reference video → pick a preset → Generate.

Availability: Input up to 2 minutes, output up to 10 seconds, 9:16 or 16:9.

Nano Banana 2 Lite is Google's fast, low-cost image model at 1K, made for rapid iterations, bulk content, and social previews before you commit to a hi-res final. Text rendering and general quality stay on par with Nano Banana 2.

• Up to 14 reference images.
• Pull a person, a scene, and a product into one composition.
• Auto aspect ratio.
• It takes framing from your reference, or pick from the full ratio set.

To get started: Image → model selector → Nano Banana 2 Lite.

Availability: 1K only. Use Nano Banana 2 when you need 2K or 4K.

Gemini Omni Flash from Google DeepMind reads text, images, and video together and composes them into one coherent shot, so you build a scene from references instead of describing it from scratch.

• Consistent characters. Face, outfit, and voice hold across every edit, so you can revise a shot without recasting it.
• Real-world physics. Gravity, weight, and collisions behave correctly, so motion looks filmed, not simulated.
• Edit by talking. Change one element of an existing clip in plain language instead of regenerating the whole scene.

To get started: Video → model selector → Gemini Omni. Add a prompt, or bring an image, start and end frame, or a clip to edit.

Availability: Preview (beta). 720p, clips of 4, 6, 8, or 10 seconds, 16:9 or 9:16, with native audio.

App Builder in Supercomputer turns a prompt into a deployed website, so you go from idea to a shareable link without touching a codebase.

• Full stack, not a mockup. Front end, back end, database, auth, file storage, and API integrations.
• Assets generated in place. Images and video are created with Higgsfield inside the site, no stock or placeholders.
• Yours to iterate. Edit meta tags and link previews, store secret keys, and remix other published sites.

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

Availability: on all paid plans. Each build and deploy uses credits to generate assets, so heavier sites use more.

The Unlimited Models Marketplace lets you 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.

• Pick your model. Nano Banana Pro, Seedance 2.0, Enhanced Seedance 2.0 Fast & Mini, GPT Image 2, Kling 3.0, or All Models.
• Pick your window. 1, 7, or 14 days, so you can match a launch or a client sprint.
• Set your quality tier. Where a model offers tiers, the top tier includes every lower resolution.

To get started: open the Unlimited models button in your profile menu, or Manage Account → Subscription → Get Unlimited. A banner and a modal on the credits page also link to it.

Availability: on Pro and Plus plans and above. Generations through MCP, CLI, Canvas, and Supercomputer still use credits. See the Pricing page for your options.

Seed Audio 1.0 from ByteDance generates speech, sound effects, and music together from a single prompt, so a clip gets voice, ambience, and a soundtrack in one step instead of three tools and a manual mix.

• Multi-speaker with continuity. Several voices hold across the scene, with emotion and accent control.
• Ambience, music, and Foley. Rain, traffic, room tone, a musical bed, footsteps, and textures.
• Context-aware. Give it text, an image, or an audio reference to guide the result.

To get started: Audio → model selector → Seed Audio. Describe the scene, including how long each element should run.

Availability: outputs in WAV, MP3, PCM, or OGG. Voice cloning supports up to 3 audio references. Length is set through your prompt, not a duration field.

Higgsfield Audio is now in MCP: generate a voiceover, create a custom voice, change a voice, and translate, right from the chat.

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

Soul Cinema generations were failing. Resolved, generations run normally again, and credits for failed generations are refunded automatically.

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

To get started: Video → model selector → Seedance 2.0 Mini. It's also part of the Seedance 2.0 Unlimited Family alongside Enhanced Seedance 2.0 Fast.

Availability: 480p and 720p, clips of 4 to 15 seconds, from text, image, video, or audio references.

Seedance 2.0 now generates video in 4K, so hero shots hold up on large screens without a separate upscale.

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

A new Standard or High bitrate choice: High gives higher video quality at the same credit cost, with a larger final file.

To get started: in Seedance, set bitrate to High before generating.

AI Employees 2.0 turns Supercomputer's agents into teammates that carry context, so "do it like last time" actually works.

• Persistent memory. Each Employee remembers your preferences across sessions.
• Connectors. Slack, Drive, Notion, Gmail, Figma, and 30 or more integrations, so an Employee can read your docs and deliver to the right folder or channel.
• Marketplace and sharing. Publish an Employee, install others in one click, or import skills and memory from Claude, Claude Code, Codex, and ChatGPT.

To get started: in Supercomputer, open an Employee card to edit its instructions, files, and memory, or create your own. Image, Video, and Text Generator are now Skills, called with /.

Generate Seedance without spending credits within the Unlimited tier at 480p and 720p. Look for the Seedance model with the Unlimited badge in the selector.

To get started: on an eligible plan, select the Seedance model marked Unlimited. Older plans may see an add-on offer for 1, 7, or 14 days.

New users get a welcome screen with three starting points, UGC video, Landing page, or Full marketing campaign, and Supercomputer sets up the right AI Employees and skills automatically.

To get started: open Supercomputer for the first time and pick a use case, or skip to start from scratch.

Kling 3.0 Turbo runs the same cinematic Kling model at much higher speed, so you lock the shot on fast drafts and render the final on standard Kling 3.0.

To get started: Video → model selector → Kling 3.0 Turbo. Supports text-to-video and image-to-video.

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

ByteDance Upscale is a new engine in the Upscale selector that raises video to 4K with frame interpolation up to 60fps in one pass, tuned for both AI-generated and real footage.

• Scene-aware presets. Match the preset to your footage, from AI generations to short series, UGC, or old-film restoration.
• Simple by tier. Pick 2K or 4K and go, without model-by-model tuning.

To get started: Video → Upscale → choose the ByteDance engine, then set resolution and preset. For fine restoration, face enhancement, or slow-mo, use the Topaz engine instead.

Generate AI Image, Realtime AI, Mockup Studio, Layer Decompose, Angles, Shots, Character Swap, Remove Background, Face Swap, Upscale, Skin Enhancer, and AI Stylist, plus Cinema Studio and Marketing Studio.

To get started: download the installer from higgsfield.ai/plugins/photoshop, then open Window, Extensions, Higgsfield. Works on Photoshop 2024 and above, and credits come from your plan balance.

Enterprise owners and admins can see members' generations across the organization, grouped by date with filtering by member.

To get started: in admin permissions → turn on View member assets, then open the Member assets section.

Higgsfield Gaming turns a description into a browser-playable game, with art, sound, music, characters, and multiplayer generated for you.

• Two ways to build. In Supercomputer, or through MCP in Claude 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: in Supercomputer, pick a genre and style and describe the game. Or add the Higgsfield connector and gaming skill in Claude.

Availability: games generate every asset with Higgsfield, so they use significantly more credits than a single generation. A cost estimate is shown before you build.

New reasoning models in the Supercomputer selector for planning and longer builds.

To get started: pick a model in the selector. Availability depends on your plan and selected mode.

Kling Motion Control, Click-to-Ad, Remove Background, Upscale to 4K, and Expand to 16:9 or 9:16, all callable from Claude.

To get started: update your Higgsfield MCP connection in Claude and call the tool by name.

Uploading reference images through Higgsfield MCP in Claude now works via an in-chat widget: upload, wait for Ready, and click Continue, for image-to-image, image-to-video, and character reference.

Video and Image Generation, Edit Video and Draw to Video from a prompt, Reframe, Remove BG, Upscale, Color Match to build a LUT, and Import straight to your Media Pool or Timeline.

To get started: install the plugin from higgsfield.ai, load a clip or reference, submit, then download or import the result.

Cinema Studio has a dedicated space with its own navigation, unified with Collab, so a whole production lives in one place instead of scattered generations.

• Projects and subfolders. Organize scenes and assets the way you'd organize a shoot.
• Characters and locations with @. Pull your elements into a Describe your scene prompt bar.
• Sections at a glance. Home, My Generations, My Elements, My Favorites, and Community Feed.

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

Enterprise owners can create sub-workspaces inside their main workspace and distribute credits from the shared pool, each with its own owner, allocation period, and limits.

To get started: Manage Account → Sub-workspaces → Add → set the name, credits, allocation period, and owner.

Generate images, video, and full builds inside Minecraft. Prompt to Build materializes structures block by block, with in-game Camera, Photo Slide, Video Projector, and Generation Archive.

To get started: install the mod → launch Minecraft → type /higgsfield auth in chat, then confirm in the browser.

Image generation across models, Vector Generation (SVG), Mockup Studio, Remove Background, Color Grading, Expand, Relight, Angles, and Face Swap. FigJam adds Canvas Nodes with chain workflows.

To get started: in Figma or FigJam → open Plugins, Find more plugins → search Higgsfield AI. Generate, then drag and drop or Insert.

Grok Video 1.5 from xAI animates a still image while keeping the source composition and style, using your image as the first frame, and it also works from text alone.

To get started: Video → model selector → Grok Video. Add a prompt, and optionally one start image.

Availability: 1 to 15 seconds, 16:9, 9:16, or 1:1.

AI Employees are ready-made agents you call with @, and they orchestrate the right skills for you, so you brief a task instead of wiring up tools.

• Ready-made roles. Video Generator, Image Generator, Cinematic Director, UGC Creator, and more.
• Marketplace. Browse Employees, Skills, and Connectors, and install with one click.
• My Office. One place for everything you've installed and created.

To get started: open Supercomputer → browse the Marketplace → call an Employee with @. Skills didn't go away, they live in the Marketplace and run with / or automatically.

Recraft v4.1 is built for professional design work, with strong typography, clean lines, and brand visuals, and it can output editable vector SVG straight from a prompt.

• Four modes. Standard for general work, Vector for logos and icons, Utility for product shots, Utility Vector for brand assets.
• Native vector. Vector modes scale without quality loss, ideal for logos and icons.
• Brand controls. A palette of up to 10 colors, a background color, and 1K or 2K output.

To get started: Image → model selector → Recraft → choose a mode.