ArtCraft: a staging IDE for AI image and video work

Who ArtCraft is built for, and who it is not

ArtCraft describes itself as "the IDE for artists" and as a tool for interactive AI image and video creation. The README frames the problem directly: artists need control over composition, character identity and camera placement, and the project's answer is to build the scene before generating it. That is a specific claim about workflow. Instead of iterating on prompts until a frame happens to look right, you place actors in an environment, pose them, set a camera, and then generate.

The intended users are named in the repository description: artists, designers and filmmakers. The feature list matches that audience rather than a general audience. Image to Location places virtual actors in a consistent environment so multiple shots can be planned in the same room. Scene Blocking with Kitbashing combines 3D asset kits to control camera angles, object placement and depth. Character Identity Transfer uses a posed mannequin to guide the placement and pose of a character. None of these are features you reach for when you want one decorative picture.

The wrong user is equally clear. If your workflow is a single prompt and a single output, ArtCraft asks you to learn a 3D staging surface first. The README acknowledges this tension by offering both paths: start with a prompt when you want to explore an idea quickly, then use the canvas and scene tools to refine it. That ordering is a hint about where the project thinks the value sits.

How the crafting engine is put together

The repository is a Rust workspace with one default member, crates/desktop/artcraft, which is the desktop application. Everything else in the members list is either a library or an API client. That layout tells you more about the architecture than the marketing copy does.

The API client crates are the interesting part. The workspace includes clients for fal, gmicloud, grok (two variants, an api client and a consumer client), kinovi web, midjourney, openai sora, and worldlabs (again in api and consumer variants), plus artcraft's own api defs, client and router crates. The README's phrase "choose the models that fit your work" is implemented as a router that sits in front of several third-party providers rather than a single bundled model. The distinction between an api client and a consumer client for the same provider suggests some integrations target an official API while others drive a web-facing consumer surface.

Surrounding that are support crates that read like a browser stack: browser_emulation, cookie_store and cookie_store_wrapper, cloudflare_mitigation, cloudflare_errors, http_headers, and a vendored tauri-plugin-http. A desktop app that needs cookie storage and Cloudflare mitigation is a desktop app talking to web endpoints that were not necessarily designed for it. That is a real design constraint, and it is the kind of thing that breaks when a provider changes its front end.

On the data side there are crates for sqlite tasks, public enums and tokens, plus a _database directory and a .sqlx directory at the top level, which points to SQLx compile-time checked queries. Local state is therefore persisted in SQLite rather than held only in memory. The frontend directory sits alongside the Rust crates, so the UI is not written in Rust.

Installing ArtCraft and planning a first shot

The README does not give a command-line install. It points to three places: getartcraft.com for a download, the GitHub releases page, and _docs/dev_setup.md for building from source. The most recent release listed is artcraft-v0.37.0, published on 2026-09-11; earlier ones include artcraft-v0.32.0 and artcraft-v0.29.0. If you want the packaged application, start at the homepage or the releases page rather than a package manager.

If you are building from source, the repository is a Cargo workspace whose default member is the desktop crate, so a workspace build is the natural entry point. The README directs you to _docs/dev_setup.md for the exact steps and prerequisites, and that file is the authority here, not this article:

git clone https://github.com/storytold/artcraft
cd artcraft
cargo build

Because the workspace uses SQLx, expect a database step to be part of the setup; the top-level _database and .sqlx directories exist for that purpose. The precise command is in _docs/dev_setup.md.

Once the app is running, the first real use is not a prompt. Pick a location, place an actor in it, and block the shot. The README's Image to Location feature is the intended starting point: place virtual actors in a consistent environment and plan multiple shots in the same room. From there, Character Posing lets you pose characters and set your camera before generating the final shot, and Character Identity Transfer uses the posed mannequin to guide placement and pose. The output of this stage is a staged scene, not an image. Generation comes after.

When you do want to generate, the README's guidance is to start with a prompt to explore an idea and then refine it with the canvas and scene tools. Background Removal and 2D Image Compositing are the tools for that refinement pass, and Mixed Asset Crafting is what lets image cutouts, worlds and 3D meshes share one scene.

Where ArtCraft breaks down

The clearest limitation is the one the README states about itself. It says more previews are coming for scene blocking, canvas editing and scene relighting. Scene relighting is listed as a future preview, which means lighting control is not presented as finished. If your work depends on matching light between a generated character and a plate, that is the gap to check first.

The second limitation is structural. The API client crates target third-party providers, and several of them are consumer-facing surfaces rather than documented APIs, which is why the workspace carries browser_emulation, cookie_store and cloudflare_mitigation. Integrations built that way are fragile by nature: they depend on a provider's web front end continuing to behave the same way. The README does not document what happens when a provider changes, and it does not describe an offline mode, so the generation features should be assumed to need network access and a working account with whichever providers you route to.

The third is the licence. The repository's licence is reported as NOASSERTION, and the README does not explain the terms. That is not something to guess at. Read LICENSE.md directly before you build anything commercial on top of it.

Finally, consider whether you are the audience at all. ArtCraft is a desktop application with a 3D staging surface. If your actual need is a batch of images from a text prompt, the staging model is overhead, and a hosted service with nothing to install will get you there faster.

ArtCraft compared with prompt-first tools

The honest alternative is the category ArtCraft is arguing against: hosted, prompt-first image and video generators. The difference is not quality of output, it is where the control lives. In a prompt-first tool, composition is an emergent property of the text you write, and consistency across shots is something you chase by reusing seeds and phrasing. In ArtCraft, composition is an explicit object. You place a mannequin, you set a camera, you reuse the same environment for a second shot. That is the whole point of Image to Location and Scene Blocking with Kitbashing.

The trade-off runs the other way too. A prompt-first tool has no workspace to build and no 3D scene to maintain. ArtCraft's workspace is a Cargo project with SQLx migrations and a Tauri-based desktop shell, and its generation path runs through a router that fronts several providers. If you are not planning multiple shots in a consistent space, you are paying the setup cost for a capability you will not use.

A second comparison is worth making inside the project itself. ArtCraft routes to multiple providers rather than shipping one model. That means the model you get is a configuration choice, and the README's "choose the models that fit your work" is a genuine design position, not a slogan. The cost is that the project inherits the reliability of every provider it integrates, and the workspace layout shows how much machinery that requires.

Maintenance, releases and what upgrading costs

The repository is not archived, and the last push was on 2026-09-14. Release cadence over the listed window is uneven: artcraft-v0.29.0 on 2026-04-23, artcraft-v0.32.0 on 2026-05-14, then artcraft-v0.37.0 on 2026-09-11. The minor version jumps between releases are large, which suggests work lands in batches rather than as a steady trickle.

For anyone tracking the project, that cadence matters more than the version number. A jump from 0.32 to 0.37 across four months means the surface can move between releases, and the workspace layout reinforces the point: there are API client crates per provider, and provider-side changes are exactly the kind of thing that forces a release. Pinning a version and reading the release notes before moving is the practical approach. The repository does not document a rollback procedure.

On licensing, the reported licence is NOASSERTION and LICENSE.md is the file to read. Nothing in the README describes commercial terms, redistribution rights, or what happens to generated output. Treat that as an open question rather than an assumption, and get your own answer from the file itself. This is not legal advice; the point is that the repository does not settle the question for you.

The upgrade cost also includes the build. A Rust workspace with SQLx compile-time checks and a vendored Tauri plugin is not a drop-in binary swap if you build from source. The README points to _docs/dev_setup.md for that reason.

Editorial conclusion

Adopt ArtCraft if you plan shots rather than one-off images: the 3D scene blocking, character posing and identity transfer features exist precisely because composition should be decided before generation. Skip it if you want a browser prompt box or a hosted service with nothing to install, and skip it if you need a documented offline mode, because the README does not describe one. Before committing, check the LICENSE.md text, read _docs/dev_setup.md for the current build requirements, and confirm on getartcraft.com which models your own account can reach.

Frequently asked questions

Official sources

1. Issues
2. Project website
3. README
4. Releases
5. storytold/artcraft on GitHub