My art should outlast the software I use to make it

A digital artwork has two forms for me. The first is the image I show: its composition, colours, light and the story someone discovers in it. The second lives inside the working file. It contains the individual elements, masks and edits that brought the image into being.

I do not want to lose that second form. I want to reopen a work later, understand it and continue developing it, even if the software I used to make it has changed or disappeared.

That is the motivation behind OAD, currently short for Open Artwork Document. It is an investigation into an open file format for layered digital art, with room for AI and new approaches to image editing. There is now a design. The important practical tests are still ahead.

My work does not fit inside one tool

My creative development began with photography. From photographs, my interests grew into photomontage, digital painting, 3D and eventually AI. I also describe that development on my art website.

Every new tool opens up different possibilities. A photograph can become part of a composite. A 3D render can provide material for a montage. A generated element can be edited, masked and incorporated into a composition that takes an entirely different direction.

Contrast and the colour red play an important role in my work. But an area of colour is more than its colour. It can guide the eye, introduce tension or counterbalance another element. Its function emerges from its place within the whole.

That is why a layer means more to me than a separate picture. It is also a choice within the artwork. I want a future working file to leave room for recording that choice, without pretending that every artistic decision can be fully captured in words.

What remains when an application disappears?

I have a substantial collection of Photoshop files. They contain work that I do not want to make dependent on the survival of one company or application.

An exported image matters: it preserves what I made. But the export alone does not necessarily let me return to an individual layer, an original source image or a mask. I retain the appearance while losing some of the ability to keep working.

OAD therefore needs to try to preserve several things together: the accepted image, its usable structure and the available information about its parts and how they came into being.

That requires openness beyond a file that can be unpacked. Another developer needs to understand what the stored information means. There need to be readable specifications, available decoders and examples from which an independent application can be built.

The test I have in mind is practical: can I open the file on another computer and recover its parts without Adobe, without my own ARtBase and without an internet connection?

The image I approve must survive

One difficulty is that applications can execute the same recipe differently. A combination of layers, filters and colour adjustments may produce a slightly different result elsewhere.

For an artwork, that matters. A small difference in light or colour may be precisely what I spent time refining.

Alongside source material and settings, I therefore want to preserve the accepted visual result. In the design, this is the authoritative render: the pixels of the version that serves as the visual reference.

That does not solve everything. A stored final image does not prove that every layer remains equally editable elsewhere. OAD needs to state honestly which parts are fully accessible, which survive only as rendered results and which operations still depend on the original software.

That honesty matters more to me than an early promise that every Photoshop file can be migrated without loss.

AI needs to understand the parts

Many AI image tools work with a flattened image. They must infer where elements begin and end and how they relate, even though some of that structure may already exist inside my working file.

My intention with OAD is therefore to allow every image-bearing component to carry a description: the complete work, but also individual layers, groups, sources and masks. A hidden layer may still contain valuable material. I do not want to skip it automatically.

A description should be able to consider both the component itself and its context. An isolated texture may look abstract, yet serve to make a background more unsettled within the composition. A mask becomes easier to understand when you can also see the image it affects.

Suppose I later ask an AI assistant to soften the texture behind the subject. I want it to be able to locate the relevant components and propose a targeted change. Whether that works reliably in practice remains to be tested.

It must also remain clear who established a piece of information. My description of my intention is different from what a model thinks it recognises. AI can be wrong. An interpretation must not silently become part of the history of my work.

Working in parallel, then continuing together

One part of the design I particularly value is branches: parallel layer stacks within the same document.

I can develop different elements or treatments within them, then bring them together at a merge. Shared operations can sit above that merge, such as a colour adjustment or texture applied to the combined result.

The original branches remain separately editable. I do not have to sacrifice their structure to continue working on the whole.

The saved document must also record which branches are active and which settings belong to a particular output. Otherwise, a future application cannot know which state produced the preserved final image.

This branch model needs dedicated test examples. An existing Photoshop file must not acquire an invented history of branches it never contained.

Preserving a painting and its instructions

I am also working on an AI-driven painting engine. That makes another question concrete: what should be preserved when an image emerges from a sequence of structured instructions?

For me, the image and those instructions belong together. In OAD, that result could live as a layer or part of a branch, alongside the original JSON recipe, its sources and the engine version.

If I change a setting, I want to be able to generate a new version without losing the previous one. The file must continue to identify which result belongs to which recipe.

But the image remains important here too. If that painting engine is no longer available, I still need to be able to recover the accepted painting. Preserving the recipe creates possibilities for later; it does not guarantee that every future application can render the same result.

I also want to preserve my original intention separately. The desire to paint a background more loosely may remain useful longer than the precise instruction language used by the engine that executes it today.

Open development means taking existing work seriously

I do not want to start from scratch for every component. OpenRaster, Krita and existing image formats already contain solutions we can learn from. Extending something that exists may ultimately prove more sensible than creating an entirely new format.

Descriptions, prompts and provenance also exist in other standards. The question is how to connect them usefully to the components of an editable artwork.

For now, OAD is growing as a standalone subproject within ARtBase, my own environment for managing and connecting my art and sources. The format must eventually be able to exist independently of it. My own software must not become another dependency that brings us back to the same question in twenty years.

First, demonstrate that it works

The next pieces of evidence are concrete. A small test document needs to show that branches, merges and shared operations can be preserved. After that, I want to migrate one complex PSD or PSB file of my own, with a clear inventory of its layers and every limitation we encounter.

The accepted final image needs to be compared with a Photoshop reference at its original precision. We also need to investigate how much editable structure survives and whether descriptions of individual components actually help. Storage requirements and processing time belong in those measurements too.

Those results do not exist yet. This article describes the direction and the reason for doing the research. A later publication should show which parts hold up.

For me, it begins with a simple wish: to keep working later on what I make today, with access to the image, its parts and the choices I have been able to preserve — and the freedom to use different tools.

---

Published 2 November 2026. Adapted into English from my Dutch manuscript. OAD is a preliminary design: complex PSD conversion, pixel-identical migration and independent recovery have not yet been demonstrated. See the OAD project overview for context.