Visual AI

Generation is becoming easy. Control isn't.

A convincing synthetic image can sometimes be generated in seconds.

Reliable production, repeatability and control are a different matter.

An AI character is easy to create. Creating the same character again tomorrow is harder.

Synthetic Identities

Our focus is stable synthetic identities.

The goal is not simply to generate an attractive AI character, but to keep the same figure recognizable across different scenes, perspectives, clothing, lighting conditions and production systems.

We combine reference material, different models, controlled workflows, automated quality checks and human selection.

Identity is not one good output. It is a reproducible state.

Synthetic character Elena in an aircraft seat, showing stable character identity in a changing environment
Synthetic identity · changing context · stable character
Synthetic character in a realistically lit yoga scene with several people
Realistic world · light · environment · production

Image & Video AI

Current image and video systems can produce impressive results — especially on short, clearly defined tasks.

A spectacular demo and a reliable production process are two different things.

The more complex a scene becomes, the more important the following become:

  • consistency
  • spatial logic
  • control
  • repeatability
  • quality assurance
  • and conventional post-production

That is why we are less interested in what a model can do in its single best attempt.

How reliably can you actually work with it?

AI-native Workflows

We do not lock ourselves into a single AI system.

Depending on the task, we combine local models, cloud-based systems, specialized AI roles, conventional software and generative tools.

One system may preserve identity well. Another may be stronger at motion or composition.

Software is often better at deterministic tasks.

AI becomes particularly interesting where context, interpretation or exploration is required.

The interesting part often lies not in one model, but between the systems.

From variations to new ideas

For us, AI is increasingly becoming more than a production tool.

It can also become part of ideation.

The interesting point is not when a system merely generates twenty variations of the same idea.

It becomes interesting when it expands the existing search space:

a new axis,
a different mechanism,
an unexpected combination,
a direction nobody had considered before.

Variation works inside a search space. Creative exploration can change it.

Selection and evaluation remain human.

But the idea no longer has to come exclusively from a human.

Synthetic character at a fuel pump pouring red wine from the nozzle into a glass as an example of an absurd visual mechanism
Mechanism test · plausible world · impossible idea

Selected Projects

Alongside talks, we selectively take on Visual AI workshops, consulting, synthetic-identity projects, AI prototypes, and the development and evaluation of suitable production systems and workflows.

We are not a conventional full-service AI agency.

We are particularly interested in tasks where something can genuinely be tested, understood or newly built.