open-ai-canvas

What it is in one line

One canvas holds a whole production: pull character assets out of a novel, then arrange storyboards, images, video, and audio into a film.

Its own line is "take a story from words to the screen" — better writing than most commercial products manage, but still a literary flourish rather than a description of what it does.

Who built it

The GitHub organization ddcat-ai. Very little else.

One thing has to be recorded: the project homepage points at ddcat.pronhubcn.com. The shape of that domain — a misspelling of a well-known adult site plus a cn suffix — is seriously suspicious. It may be a developer's throwaway server. It may be something else.

Unresolved, and it needs checking. Until it is, don't cite this project anywhere formal, and don't upload material to that domain.

What it actually does

• Feed in chapter text → it works out the characters and the structure of the plot
• Freeze a character into an asset → look, voice, and temperament bound into one reusable object that every later generation step references, which is how you stop "the same character looks different in all ten images"
• Arrange it on an infinite canvas → storyboards, images, video, and audio organized in one space instead of hauled between a dozen tools and folders
• Wire in multimodal generation → call external generators for images, video, and voice from inside the canvas
• Agent workflows → chain those steps into an automated run

What it deliberately does not do: it trains and serves no models of its own. It is a container; every generative capability comes from an external API. That is its biggest lever and its biggest fragility at the same time.

What old behavior it replaces

The real AI filmmaking workflow today is: write the outline in a chat model, generate stills in one image tool, generate clips in a video tool, stitch it in an editor, and move material between all of them through docs and chat apps.

The pain isn't in any single tool. It's in the hauling between them: a character's face doesn't match across tools, the shot order lives in someone's head, and the assets are scattered across a dozen places. This is what open-ai-canvas is replacing, and the old behavior is unmistakable.

Business model

AGPL-3.0. That is not a casual pick. AGPL requires anyone who offers a network service based on the code to open-source their version too. It is the classic setup move for dual licensing: stop the big platforms taking it for free, push commercial users toward buying a license.

Read: choosing AGPL says the author has thought about revenue, and a commercial license or hosted edition is probably coming. There is no pricing yet — this is still the accumulate-stars phase.

Hard numbers

• Open-source traction: 535
• Version v1.0.43 — the version number says it iterates often, not a toy thrown over the wall
• Users, revenue: undisclosed

Four-way read

The call

This is a bet on the container, not the generator.

It builds no model. It builds the canvas that organizes what the models produce. The logic underneath: model quality keeps rising and converging, so what's left to compete on is who owns the structure of the work — the relationships between characters, shots, and assets — not whether any single frame is pretty.

The transferable rule: when upstream capability commoditizes fast, value migrates to the structure that organizes that capability. This is the same move a model gateway makes one layer down. It's just happening at the creative layer instead.

The cost is that this kind of product is brutally hard to run: it has to stay wired into a pile of fast-moving generation APIs, and any one of them changing an interface breaks a link. It also has no moat — a canvas is easy to copy, so the only real barrier is the data structure behind asset organization, and it isn't obvious yet how deep that goes.

It and shuohao-skills are two bets on the same judgment: one on the upstream (novel into material), one on the container (material into a film).

What to watch next

1. What ddcat.pronhubcn.com actually is — nothing else matters until that's answered
2. Whether a commercial license or hosted edition appears — AGPL is the setup; watch for the payoff
3. Whether the character asset schema stabilizes — refactoring it every release would mean the author hasn't settled it either

What you can take from it

Go read the character asset schema. How it defines a character's look, voice, and temperament, and keeps those consistent across separate generation steps, is the hardest problem in any multi-step creative pipeline — and someone has published an answer.

Pricing: AGPL plus a commercial license is a ready-made template for anyone who wants open source to do their marketing without handing the product to a competitor.

Verdict

Strong pick, but check that domain first. As a public reference for how to organize a multi-step creative workflow, reading its code beats reading ten write-ups.