Peer, can you tell us, why this project was approached with AI in the first place?

In the first place it was for budget reasons. Since the scene, a woman emerging from a piece of furniture, would have been nearly impossible to realize using traditional production methods, a conventional approach would have required specially prepared products to allow the actresses to exit in an aesthetically pleasing way. In addition, this would have involved a significant amount of extra compositing work to remove visible support elements at the bottom of the chair, as well as extensive cleanup and error correction. All of this naturally involves costs.

Tell us a little bit about how AI was built into the process.

We developed our own AI process that served as a creative framework throughout the entire production. At the same time, we fully embraced the fact that AI operates far more dynamically than traditional pipelines and therefore requires a high level of flexibility.

Nevertheless, core principles of classic production remain essential: moving images in AI-driven production should never be generated without a clear framework. Especially in a film that relies on strong character consistency, defining casting and styling upfront is fundamental even, and especially, within an AI-based workflow.

Production reality also requires alignment and clear decision-making. Although this was an AI-driven project, we held a traditional PPM to align all stakeholders early on. Within the process, we deliberately defined several points of no return, not to limit creativity, but to clearly mark decisions as approved and avoid reopening topics that had already been signed off, which in AI production can be a killer.

What were the biggest changes people had to deal with in the process due to the use of AI?

The biggest shift and at the same time key challenge was understanding that we no longer have the same level of control as on a traditional film set. It is not possible to brief an AI in such a detailed way that it delivers the exact performance one might envision or be able to achieve on set. This is not inherently negative, but it does require a conscious adjustment in mindset.

The process can be compared to a choreographer working with a dance group: he is allowed to read out a description and show two reference images, after which the choreography is performed. If the result is not satisfactory, the text and images can be adjusted, but on the next attempt, the dance group has completely forgotten everything from the previous run.

We are rooted in a production process that has evolved and proven itself over decades. Transitioning to a new, AI-driven workflow is therefore not trivial, but a fundamental shift that requires creative, structural and technical rethinking.

Let us know what where the biggest learnings from that.

The biggest learnings from this project were rooted in accepting the actual capabilities of AI while remaining flexible in terms of output. AI unlocks tremendous creative potential, but it also requires a shift in mindset: results are less precisely controllable than in traditional production, and this needs to be clearly communicated both, internally and to the client.

Equally important was continuing to rely on established expertise across departments. Working with AI does not replace the knowledge of disciplines such as styling or art direction. Decisions around outfits, proportions, and overall look still need to be made by experienced professionals, rather than relying solely on AI-generated suggestions.

Another key learning was the importance of a clearly defined AI process. Without this structure, the production would have become far more chaotic and inefficient. The process helped manage complexity, align expectations, and establish clear, binding decisions.

Finally, while AI creates the impression that almost anything is possible instantly, reality proves otherwise. Strong results still depend on highly complex workflows, coordination, and iteration. AI can accelerate production, but it does not eliminate the need for process. Respecting that process is essential to unlocking its full potential.

Tell us a little bit about the models and the pipeline.

It depended on the specific task. Our goal was to keep as many steps as possible within our custom ComfyUI workflows to maintain maximum control and achieve the best possible output. Since the market is currently dominated by differently trained models, continuous testing was essential: Where do certain requirements work best, and which setup delivers the desired output? The decisive factor was always the balance between control, consistency, and creative freedom. Accordingly, we combined models and switched manually between tools as needed to leverage their strengths.

If you can share: which specific tools/models were used, and for what tasks?

To many to tell. ComfyUI was one of them. Some locally, some not.

Have there been any concerns about legal issues?

At present, there are few areas in the media industry that are as legally undefined as working with AI. For this reason, it was particularly important for us to approach this uncertainty responsibly. E.g. by not using any copyrighted material as input, we deliberately ensured that we did not push the boundaries of this legal grey area.

How about the total timeline and team setup?

The production ran for approximately six weeks. The core team included two to three AI artists handling concept development and image and video generation, supported by two to three compositing artists focused on cleanup, stabilization, and visual refinement. The setup was completed by a traditional post-production pipeline including color grading, sound design, and producing.

Where any assets or systems made reusable?

In General ComfyUI Workflows for different kind of tasks

What deliverables were produced? (hero film, cutdowns, social assets, stills, localizations)

Hero Film + Stills

How do you think can the approach of this project be seen as a blueprint, and how do you think AI will change commercial filmmaking in the future?

How the market will develop over the coming months remains to be seen. We use our process as a blueprint and continuously refine it across projects. By incorporating ongoing learnings, we have steadily improved the workflow and achieved consistently strong results.

Peer, thanks for the Interview!

Client: MöbelPfister AG