FractalPerception.com

Exploring Aesthetic Perception in High-Dimensional Parametric Fractal Systems

1. Making the creative process observable to AI — and enabling AI to participate in it

FractalPerception is an independent research program at the intersection of artistic creation, parametric fractal systems, aesthetic perception and artificial intelligence.

The project is based on a corpus built since 2020 from the artistic research of Philippe Chevalier / Philoxerax. Its distinctive feature is that the images are available not only as pixels. They remain connected to their generative structure — mathematical formulas, parameters, transformations, layers, gradients and blending modes.

This dual access, visual and parametric, makes it possible to connect several levels that are usually considered separately:

generative structure ↔ visual rendering ↔ perception ↔ aesthetic decision

The images are also non-figurative and non-referential. Analysis can therefore focus directly on their visual structures — forms, rhythms, densities, colors and balances — without first depending on the recognition of a represented subject.

2. An existing parametric corpus

The corpus currently includes:

• 376 fractal compositions
• 116 unique functions
• 34,369 composition-level parameters
• 1,396 shapes — raw parametric forms used as the basis for layers within the compositions
• a high-resolution bitmap rendering for each image

This initial material comes from artistic research on fractals begun in 2020 and based exclusively on a process of creation and aesthetic selection. It predates FractalPerception and served as the project’s starting point.

3. What exploration with AI already makes possible

The system gives an AI access not only to the images in the corpus, but also to the parametric files and formulas that underlie them. It also allows the AI to act on this parametric space to produce new proposals.

Analyze

An AI can read the parametric structure of a work, identify the formulas used, compare their parameters with the rendered image, and search for recurring patterns in aesthetic choices.

The objective is to test whether repeated decisions — what is kept, modified or rejected — reveal patterns coherent enough to contribute to describing an aesthetic grammar.

This first exploration is documented on the → Exploration page.

Explore and co-create

The channel also works in the other direction: the AI can directly generate parametric files that can be opened by the creation software.

Following the artist’s instructions, it can explore a formula or a region of parameter space at scale and produce dozens or hundreds of variants. The artist evaluates the results visually and decides which direction the next explorations should take.

AI thus becomes an instrument for parametric exploration and analysis within the creative process, while aesthetic validation and direction remain human.

This experiment is documented on the → Co-creation page.

4. A bidirectional channel beyond words

The aesthetic dialogue no longer relies solely on language and prompts. The AI acts directly on parametric files to generate new proposals; the selections, transformations and rejections made visually by the artist become, in turn, data that the AI can analyze.

Exploring shapes

The channel operates as a loop:

the artist starts from a formula and first explores it in the software
→ the artist settles on an initial shape considered worth exploring further
→ the AI reads this file and analyzes the corresponding parameters
→ following the artist’s instructions, it generates a batch of new parametric proposals
→ the software makes them visually accessible to the artist
→ the artist selects, rejects, qualifies or transforms the results
→ the artist then decides the direction of the next batch through the same channel
→ the AI generates new proposals based on that direction.
Progressively, the shapes become the layers assembled by the artist to build the composition.

Composition

Beyond the creation of shapes, where AI participates directly in the creative process, it can already observe in detail a second process conducted in parallel by the artist: composition. It accesses this process through the files and successive saved states of the composition, including coloring, layer superpositions and the inks associated with the layers.

In the longer term, it is reasonable to envision the AI also intervening in this process by proposing transformations, layer combinations or new directions for the composition.

A black box turned into an interface

The software used to build the corpus is old, and its rendering engine largely functions as a black box: the parameters supplied to it are accessible, but its internal process for transforming those parameters into pixels is not fully observable.

This limitation led to a practical solution: the parametric file becomes the interface with the AI, while the existing software continues to provide the visual rendering. There is therefore no need to rebuild the creation engine in order for AI to participate in the process.

artist’s creation ↔ software visual rendering ↔ parametric save ↔ AI analysis

The remaining gap between parametric description and actual rendering is also a potential object of study.

5. From an image corpus to a corpus of the creative process

This bidirectional channel now makes it possible to preserve information that usually disappears when only the final work is archived:

• exploratory variants;
• proposals that are retained or rejected;
• successive versions;
• forks and abandoned branches;
• layer additions, modifications and removals;
• the lineage of a shape from its exploration to its integration into one or more compositions;
• aesthetic decisions between two states.

A fractal is therefore no longer described only as a final state of parameters in a file, but as a parametric genealogy of decisions.

The process is documented through the files and successive saved states, without attempting to continuously track every manipulation performed inside the software.

FractalPerception can thus evolve from an image corpus into a corpus of the creative process.

6. The current challenge: structuring the corpus

The current work is to design a model capable of preserving, in a coherent and usable form, for each composition and for the different states in its genealogy:

mathematical and parametric data
formulas, variables, positions, layers, blending inks, gradients;

the rendering and its description
image, visual analysis, semantic data;

the process
versions, retained or abandoned branches, progressions, comments and decisions.

A relational database must serve as the source of truth in which all this information is stored. A JSON / JSONL export must make it possible to extract, for each composition, all the information associated with it in a form that can be used by AI systems.

The model is currently being tested on the genealogy of Fractal 154 / God Face, which includes several successive versions, an abandoned branch and layers resulting from co-creation. This analysis should make it possible to validate the structure before extending it to the existing corpus and future compositions.

Partners sought

The technical feasibility of the bidirectional channel has been demonstrated through real-world cases. This channel connects the AI and the artist throughout the creative process via the parametric space. On this basis, FractalPerception is now seeking to establish two parallel partnerships: one institutional and artistic, the other scientific.

Institutional and artistic partnership

The objective is to make the birth of an artwork visible to the public through the dialogue between the artist and the AI, together with the different traces produced during its creation: parametric proposals, selections, rejections, transformations, successive states and audiovisual documentation.

FractalPerception is seeking an institution or curator with whom to devise the most appropriate way to present this material to the public.

This partnership may involve:

• art centers and venues dedicated to digital art or emerging forms of creation;
• foundations;
• art × technology / art-science institutions;
• curators and arts programmers.

Learn more about the creative process as exhibition material

Scientific partnership

The objective is to transform this system and the existing material into a structured, documented and methodologically robust corpus, enabling AI to learn how aesthetic judgment is formed.

FractalPerception is therefore looking for laboratories, academic teams or R&D teams working notably on artificial intelligence and:

• visual perception;
• computational aesthetics;
• generative systems;
• human–machine interaction.