How BITT Animation Runs Hybrid AI Production on Higgsfield with Cinema Studio and Supercomputer

BITT Animation has been producing animation and VFX for 24 years. The studio was founded by Franco Bittolo, CEO and Director, and built up alongside Cristian Morales, COO and Director, the first partner to join, years before AI entered the picture.

Its client list includes Coca-Cola, PepsiCo, Toyota, Ford, Volkswagen, Unilever, Arcor, Nestlé, Nivea, Burger King, Spotify, Renault, Quilmes, Nissin Cup Noodles, Topo Chico, Mondelez, KFC and Honda, with agencies including Ogilvy, VML, TBWA, BBDO, GUT, Publicis, Media.Monks, DAVID, Le Pub, DraftLine and Isla.

The work runs from campaign films to series contributions, among them El Eternauta, alongside collaborations with Oscar winning director Armando Bó and with international partners including The Mill, Untold and 1st Avenue Machine.

Close to five years ago, Gonzalo Canepa joined the company as a partner to build a new department focused on emerging technologies. That department became BITT Plus, the studio's AI branch, and what began as an exploration has gradually reshaped how the whole studio produces: today at least 70% of the work moving through it involves AI at some stage. In this conversation, Gonzalo, Chief Creative Officer and Chief AI Officer, explains how a studio built on Maya, Houdini and Nuke arrived at generative production years before most of its peers, how a plate actually gets through a hybrid VFX pipeline, and why the change that mattered most was one nobody at the studio predicted. Who is BITT Animation, and what does having two brands under one roof let you do?

BITT is two connected brands in the same house. BITT Animation is the traditional animation and VFX studio, with 24 years of production behind it. BITT Plus is the AI native arm built alongside it: same team, same standards, applied to generative production.

The arrangement is a practical one. Having both brands in the same building means techniques can be mixed freely, and a lot of the studio's current work combines traditional VFX with AI generation inside a single pipeline, deciding shot by shot which approach gets the frame the job needs.

Gonzalo is direct about where the studio's advantage actually sits, and it isn't the tooling.

“We have more than 15 full time artists, most with around 10 years in the industry, people who spent their careers doing this work traditionally and who we've since trained on AI so they could keep doing it in a world that changed underneath them. That's an unusual thing for a studio to have, and it's the reason our AI output looks the way it does.”

“The judgement is trained, not prompted”

Alongside the production teams, the studio runs BITT LAB, an internal R&D group that builds pipeline tools and client facing platforms. Its standing brief is to stay ahead of what's arriving: new models and platforms get tested inside a live pipeline within days of release, which is how the studio separates what holds up in production from what only performs in a demo. Higgsfield entered the studio through exactly that process.

The team works between Buenos Aires and Spain. Gonzalo is Argentine, based in Palma, with close to 20 years in high end visual production across VFX, animation, games and real time work, and the last six focused on AI. His role bridges both titles: researching what is arriving, deciding how and where AI fits each job, and feeding what holds up to BITT LAB, where it becomes the workflows the artists use.

What did a deliverable cost under the traditional pipeline?

The studio's conventional stack is the standard high end setup: Maya, Houdini and Blender for 3D, animation and simulation, Arnold for rendering, Nuke and After Effects for comp, DaVinci Resolve for grade, Unreal Engine for real time and virtual production, with a layer of internal pipeline tooling on top. All of it is still in the building, and on certain jobs it remains the right answer.

The arithmetic on a single deliverable was the thing that changed. A 30 second spot with five characters meant months of animation from a large team, simulation passes, and render time on a farm before anyone saw a finished frame. Pre-production alone was heavy: concept artists on boards and key art, then an animatic or a previz built in 3D, which is weeks of work before a client has anything to react to.

Stills from the showreel. All produced with Higgsfield

Ask the studio to name the single biggest bottleneck and the answer comes back as two that turn out to be one.

“Time and workforce, and the two were the same problem. Traditional production doesn't scale by working harder, it scales by adding people, and a project's timeline was really a function of how many animators, simulation artists and compositors you could put on it. Months on simulation alone, before render time. That's the constraint that set what we could take on and what we had to decline.”

How did a traditional VFX studio end up in generative production this early?

Not through generation, as it turns out. BITT's route into AI began with machine learning for face replacement, running on live client work before the current generation of models existed. That pulled the studio into the early stages of the field, and it moved into diffusion as diffusion arrived: local Stable Diffusion, ControlNet, AnimateDiff, the whole open source experimental period.

BITT was among the first studios in Latin America putting AI in front of real clients on real campaigns, and much of that period was spent finding out what survived contact with a brand approval process, which is a different test from looking impressive in isolation.

Stills from the showreel. All produced with Higgsfield

The testing never stopped, because that is BITT LAB's job. Over time the pieces that held up became part of the daily workflow: ComfyUI as the backbone for controlled generation, with the strongest open models available at each point, WAN, Flux and Qwen among them, then the closed video and image models as they matured, Veo 3.1 and Nano Banana alongside others. Beyond that the studio has tested essentially everything that reached the market, Runway, Luma, Sora, Kling, Krea, OpenArt and Flow among them.

That history is the context Gonzalo puts around how Higgsfield entered the picture, and he is careful not to overstate it.

“Higgsfield came in roughly eight months to a year ago and changed the shape of the day. Not because it did something no other tool could, but because it put a large part of what we were doing across five different platforms into one place, organized, with proper control over it”

It was not the first platform the studio tried. It was the one that held up across enough of the workflow to be worth standardising a team on. The studio has since trained the whole team on it.

Where does Higgsfield sit in the pipeline?

Across most of it, and differently at each stage. The full image and video suites are the daily foundation, driven by the studio's own prompt systems, developed internally and refined across projects rather than written fresh each time. The studio is specific about the division of labour there: the platform supplies the models and the control, and what BITT brings to it is a body of prompt work built up over years of client production.

Cinema Studio is the tool the team is most attached to. Its primary use is previsualisation with clients, reaching something that reads as real direction rather than an approximation, fast enough to have the conversation in the same week. On some projects it doesn't stop at previz, and material generated there has carried through to final delivery.

The studio also works through Seedance and Gemini Omni, because a significant share of its work is VFX on existing film footage, and much of that is now generative rather than conventional.

Include video with a bug creature

How is the Supercomputer utilized in the pipeline?

Beyond the core suites, the studio has used Marketing Studio and tested Supercomputer along with the agentic workflows around it, for tasks that need to run at volume or unattended. It also makes heavy use of MCP and CLI access, connected to Claude and GPT, wiring Higgsfield into BITT's own systems and automations rather than working purely through the interface. For a studio that builds its own tooling, that is a significant part of the appeal. Recent work on breaking generated images into editable layers is currently being put through its paces with a view to how it fits the existing pipeline.

Stills from the showreel. All produced with Higgsfield

How does a hybrid VFX shot actually get made?

This is the part BITT wants described properly, because the assumption from outside is that a shot is either fully traditional or fully AI, and in BITT's work it is neither.

The original plate goes into Seedance and is manipulated directly: dropping in an animated character, replacing a moving car, changing a background, swapping a product, adding an element that was never there on the day. What comes back is a generated pass, not a finished shot.

That pass then moves into a second compositing stage, where it is combined back against the original footage. This is where the shot is actually made: reconciling the two, holding what has to stay untouched from the plate, integrating what has been generated so it sits correctly in frame. From there a final beauty pass, enhancement and upscale, using tools also inside Higgsfield, and that is what goes out before grade.

Gonzalo treats that sequence as the clearest description of how the studio works in general.

“The generative step does the part that used to consume the most hours. The traditional craft around it, plate management, comp, finishing, is what makes the result hold up as a shot rather than a clip. Neither half gets there alone.”

The same logic governs where the platform replaces something and where it adds to it. At the front of the process it genuinely replaced things: the old route to a client presentation ran through concept artists and 3D previz over weeks, and a large share of that is now built directly, with the traditional route used when a specific job calls for it rather than by default. Further down the pipeline it augments instead. The 3D and comp stack hasn't gone anywhere, and on much of the work the two run together, generated material moving into Nuke or After Effects, 3D elements feeding back the other way. Which side of that line a shot falls on is decided shot by shot. The knock-on effect the team rates highest is the least cinematic one: with the work in a single organised place, it became realistic to train the whole team on one system, instead of leaving knowledge with whoever specialised in a given tool.

What does the AI-led model change in time and cost?

Worth stating plainly before the numbers: the ranges below are properties of BITT's AI-led production model as a whole, across the full pipeline, and are not attributable to any single tool. They move with how much of a given project can genuinely go generative.

On time, Gonzalo declines to give a single figure on the grounds that it would mislead. The honest range is that work which used to take three months now lands somewhere between three and six weeks, and on certain jobs closer to a week. Where a project falls in that range depends on the brief, how much of it can go generative, and what the client actually needs. The clearest structural change sits at the front: getting to a client presentation used to mean concept artists and 3D previz over a period of weeks. That is now days, often less.

On cost, the studio can't publish figures tied to a specific client, but the shape is consistent.

“A project that would traditionally have carried a budget in the low hundreds of thousands can come in at roughly 15 to 20% of that when it runs almost entirely generative. On hybrid work, where a meaningful share stays in the traditional pipeline, it lands closer to 30 to 40%.”

Higgsfield's specific role within that, in the studio's framing, is the consolidation: running a workflow through one platform instead of five or six is part of why the model is sustainable at the studio's current scale, and not just achievable on a showcase project.

The proof point the studio returns to is a retainer with a large consumer packaged goods company, running at around 95% AI output. It runs as a continuous monthly volume, not a one off, and the speed difference against traditional production is a different order of magnitude. Elsewhere, recent AI-led work includes a ten plus spot campaign for Saladix with Arcor, a Toyota TV spot, and a music video for Bizarrap.

Stills from the showreel. All produced with Higgsfield

The change nobody predicted

“We now spend more time waiting for client feedback than generating the animation they're giving feedback on. The bottleneck moved out of the studio”

Ask Gonzalo which result matters most internally and he doesn't name time or cost. He names capacity, and the reason is a scheduling quirk that only exists in generative work.

Traditional production is continuous in a way that locks people up. A project ran for three months with the full team on it, and an artist assigned to that project was on that project, because the work never paused long enough for them to be anywhere else. One client, one team, one window.

Generative work has natural gaps in it: generation time, internal review, client feedback. Those gaps turn out to be long enough to be useful. Which suggests the old constraint was never really the hours. It was that the hours arrived without pauses in them, and a person with no pauses can only be in one place. The same artist can now move between projects during them, sometimes doing a comparable task on another job, sometimes something different depending on what they're strong at. Where the studio could previously run one major project at a time, it now runs around six in parallel, sometimes more.

Gonzalo doesn't frame the result as a throughput gain.

“That's a change to what kind of studio we can be. It's the difference between declining work because the team is committed and being able to say yes.”