ArtCraft is a desktop application for AI image and video creation that has you stage the shot in 2D and 3D before any model renders a frame; the project's own one-line pitch is "the IDE for artists." It is written in Rust and aimed at artists, designers and filmmakers who care about composition, character identity and camera placement — people who would rather build a scene than retype a prompt until something close enough comes out.
What it does
The core idea is that generation comes last. The README describes composing in 2D, staging scenes in 3D, and choosing the models that fit the work, then generating from what you built.
The demo the project links from its README walks through that order:
- placing the actor in the scene
- setting the camera
- posing the character
- turning a flat image into a 3D mesh and rotating it into position
- generating only after the scene is arranged, guided by that arrangement
The claim attached to this is the project's own: its visual tools aim for more precise, repeatable results than plain prompting. That is the authors' stated goal and what their two-minute demo shows — not an independent measurement. No one here has benchmarked output quality or shot-to-shot consistency, so read it as a design intent rather than a proven number.
How it works
ArtCraft is a compiled desktop program rather than a hosted site. The README offers a download, tagged GitHub releases, and a build-from-source path, and Rust is the main language. The 3D staging layer is where the actual work happens: the scene you block out becomes the guide for the generated frame, instead of a sentence of text carrying all the intent.
The models are deliberately not part of the package. The channel's summary is blunt about it — you bring and connect your own AI image model, because ArtCraft does not generate on its own. Treat the app as the staging tool and front end; the generation happens in whatever model you wire up.
The repository is not new. GitHub dates it to March 2022, shows commits still landing in October 2026, and counts 9,721 stars and 1,436 forks. The README also links a Discord server, a YouTube channel, X and LinkedIn, which is where the project's day-to-day conversation appears to live.
Getting started
What it needs to run is the honest gap in the material. Nothing in the saved README or the repository facts names an operating-system list, a minimum GPU, a VRAM figure or a memory requirement, so don't assume it runs on any laptop — the project doesn't list hardware requirements, and the download page is the place to check before you plan around it.
One requirement is clear, though: a model. The application is free to download and the source is in the repository, with no subscription mentioned anywhere in the material, but the intelligence is yours to supply. That means either a local image model running on your own hardware, or a paid cloud model whose API key you connect and whose usage you pay for. Nothing here suggests ArtCraft produces images without one.
There are three ways in:
- the download at getartcraft.com
- the Releases page of the repository, github.com/storytold/artcraft
- building from source by following the
_docs/dev_setup.mdguide in the repository, which is the route if you already have Rust tooling set up
The material gives no single install command, so the packaged download is the normal path for most people.
When to use it / when not
The main catch is the bring-your-own-model requirement: before your first image you have to pick a model, connect it, and accept its cost, which is more setup than a hosted generator asks of you.
Use it instead of typing prompts into a hosted generator like Midjourney when the shot has to hit a specific camera angle or pose, or when the same character has to hold up across several frames and you need the staging to be repeatable rather than re-rolled.
Don't use it if you want one sentence in and one picture out with no configuration; if you need a browser tool on a machine you don't control; or if you need to know up front that it will run on your particular GPU, which this material cannot tell you.
ArtCraft is worth taking seriously if you already storyboard, block shots or think in camera terms, and find text prompting too lossy for that kind of work — the 2D-to-3D staging flow is a real answer to the "I can't get the angle I want" problem, and the project is long-running and actively pushed rather than a weekend demo. If you mostly want fast one-off images, or you need a documented hardware floor and a bundled model before you commit an afternoon, this is not yet the tool to download; watch it instead, and let someone else report what it costs to feed.