Open Source Project #243: ArtCraft — An AI Creation IDE for Artists: Build the Scene First, Then Generate, Turning Prompt-Gambling into Craft

A desktop AI image and video creation IDE written in Rust + Tauri: composite in 2D, stage scenes and character poses in 3D, then pick a model to render. Aggregates 62 models (Nano Banana, GPT Image, Seedance, Kling, Veo, Suno, Hunyuan 3D and more) plus Grok, Midjourney, Sora, and World Labs entry points. Dual-licensed MIT/Apache-2.0, 34k+ Stars.

·8 min read·AI Tools
open-sourceai-imageai-videofilmmakingrusttauri3dcreative-tools

Introduction

"The IDE for artists."

This is article #243 in the "One Open Source Project a Day" series. Today's project is ArtCraft — a desktop app for AI image and video creation aimed at artists, designers, and filmmakers, built by the storytold team, with 34,933 Stars, 5,523 Forks, written in Rust, and dual-licensed MIT / Apache-2.0.

Most AI image and video tools interact the same way: write a prompt, pull the lever. If you want the same character in the same room from three different camera angles, or a specific pose, you rewrite the prompt and hope. ArtCraft turns that from prompting into crafting: set up the composition, character poses, and camera in a 2D canvas and 3D scene first, then hand it to a model to generate. Its roadmap names this the number-one goal — to be better than model-aggregation websites.

What You'll Learn

  • ArtCraft's "build the scene, then generate" workflow: canvas compositing, 3D staging, character posing, image-to-3D mesh
  • Which models and third-party entry points it aggregates, and how to read the 62-model catalog
  • Quick start: downloading the app and building from source (Rust + Tauri)
  • A current-state detail worth knowing: the desktop app's relationship to hosted services, and the roadmap goal of removing that dependence

Prerequisites

  • Basic experience with AI image/video generation (text-to-image, image-to-video)
  • Familiarity with 3D scene concepts like camera, placement, and depth makes the selling points easier to follow
  • To build from source, a working Rust, Node.js, and Tauri CLI setup

Project Background

Overview

ArtCraft calls itself "the IDE for interactive AI image and video creation": composite in 2D, stage in 3D, and choose the model that fits. It is not another model-aggregation website — it's a desktop app that connects visual, repeatable tools (canvas, 3D scenes, background removal, mask-based inpainting) to a range of generation models.

Author / Team

  • Developed by: storytold (the ArtCraft Team), who also run a content studio; the roadmap mentions narrative shorts and at least one feature-length AI film by year's end
  • Primary language: Rust (a Rust / Tauri desktop app with a Vite and React frontend)
  • License: MIT or Apache-2.0, at your option; third-party material is covered in the repo's NOTICE
  • Repo created: 2022-03-31 (the repository's history predates the current product shape)
  • Status: very active, with the latest push landing today; latest release artcraft-v0.41.0 (2026-09-26)

Project Stats

  • ⭐ GitHub Stars: 34,933+
  • 🍴 Forks: 5,523+
  • 📄 License: MIT / Apache-2.0 dual license
  • 👥 Contributors: echelon (20 commits), bflatastic (6), and a few others — commits are heavily concentrated in a small group
  • 📬 Open Issues: 475
  • 🏷️ Topics: 3d-graphics, ai, aivideo, filmmaking, imagegeneration
  • 🌐 Website: getartcraft.com

Quick Start

Download

Build From Source

It's a Rust / Tauri app. You need Rust, Node.js (the docs say v24.13.0 works; the launcher requires 20+), and the Tauri CLI (tauri-cli 2.10.0 works per the docs).

# Mac / Linux: start the frontend and Rust/Tauri together in one terminal
./script/artcraft/unix_dev.sh
 
# Optional: choose the starting port (it searches upward from 5193 by default)
ARTCRAFT_DEV_PORT=6200 ./script/artcraft/unix_dev.sh

On Windows, run windows_frontend_dev.ps1 and windows_rust_dev.ps1 separately. Frontend edits use Vite hot reload; Rust edits are rebuilt and the app restarted by cargo tauri dev. To type-check Rust without a database server:

SQLX_OFFLINE=true cargo check -p artcraft

Core Deep-Dive

"Build the Scene, Then Generate": Taking Control Back From the Prompt

ArtCraft's most distinctive part is the "Show, Don't Tell" feature set in its README. What they share: constrain the image visually before generating.

FeatureProblem it solves
Image to LocationPlace virtual actors in a consistent environment and plan multiple shots in one room
3D / 2D Image CompositingLayer backdrops, foregrounds, and props to build compositions with depth
Image to 3D MeshTurn an image into a 3D object you can position, rotate, and frame
Character PosingPose characters and set the camera before generating the final shot
Scene Blocking with KitbashingCombine 3D asset kits to control angles, placement, and depth
Character Identity TransferUse a posed mannequin to guide where and how your character appears
Mixed Asset CraftingCombine image cutouts, worlds, and 3D meshes in one scene

This suits filmmakers well: narrative shorts need the same character in the same set from different angles, which is exactly what plain text-to-image controls worst. ArtCraft's answer is to make the scene itself an editable intermediate artifact, instead of resting consistency entirely on the prompt.

There's also a fast lane: text-to-image, instruction-style editing with models like Nano Banana Pro and GPT Image, canvas editing with brushes, masks, and inpainting, and image-to-video — useful for exploring an idea quickly before going back to the canvas and scene tools to refine.

Models and Entry Points: A 62-Model Own Catalog Plus Four Third Parties

Supported providers fall into two groups. The first is ArtCraft's own catalog, whose collapsed README section lists 62 models:

  • 16 image models: Nano Banana, GPT Image, FLUX, Seedream families
  • 25 video models: Seedance, Kling, Veo, Sora, Vidu, MiniMax, Flux 3, and more
  • 5 music and sound models: Suno variants, Seed Audio
  • 11 3D mesh models: Hunyuan 3D, Tripo3D, Meshy, Rodin
  • 5 world and Gaussian-splat models: Marble, TripoSplat

The second group is direct third-party integrations: Grok (Grok Imagine / Video), Midjourney, Sora (Sora 1/2 and GPT Image 1), and World Labs (Marble world generation). Planned additions include Kling, Google, Runway, and Luma, plus support for existing OpenArt and Freepik aggregator subscriptions.

Two details when reading that list: many catalog entries carry a † (currently disabled — e.g. Veo 2/3, Sora 2, several Kling versions) or a ‡ (restricted, not available in the desktop app). So "62" is the catalog size in the repo's config, not the number you can use right after install.

A Candid Roadmap Line: Dependence on Hosted Services

The README says little about architecture, but the ROADMAP has a telling line: "Remove dependence on ArtCraft hosted services." Combined with the dev docs noting that backend services and website builds live in a separate artcraft-services repository, a realistic reading emerges: the desktop client is open source, but some of its capabilities currently depend on backend services hosted by the team, and the dev docs even discuss first-party API handling for login and media loading.

That's not a flaw — it's a premise to know before adopting. If you need fully offline, fully self-hosted, that isn't there today, and the team lists it as a 2026 architecture goal. The roadmap's other items — more Rust and speed, more tests and CI, and purpose-built models for relighting scenes and reposing characters — also signal a product still taking shape.

Resources

Summary

ArtCraft targets a widely overlooked problem in AI generation tools: when pulling the lever is the only interaction, creators can hardly produce consistent, controllable, repeatable work. It brings the "previs" steps of traditional creation — 2D compositing, 3D staging, character posing — back into the AI workflow.

Three things worth noting:

The scene is first-class; the prompt comes second. Place a mannequin, set the camera, and split layers before a model renders. That's a direct answer to the cinematic problem of consistency, and it's the most essential difference between this and a plain model-aggregation website.

The aggregation is broad, but check real availability. 62 models plus four third-party entry points cover image, video, music, 3D, and world generation, but a notable share are marked disabled or restricted; what you can actually use is whatever the app exposes.

An open-source client is not the same as fully self-hosted. The roadmap lists removing dependence on hosted services as a goal, which means the dependence still exists. Before putting it into a workflow with data-compliance requirements, confirm which steps go through the team's services. The 475 open issues and concentrated commit structure also suggest it's still in a fast-iteration phase.

If you're making AI shorts or storyboard previsualization, or need visual content where characters and sets stay consistent, ArtCraft is worth downloading a stable build to try.

Explore PrimeSkills — a curated marketplace of AI agents and skills, each validated against real enterprise workflows. No hype, just what actually works.

Visit my personal site for more insights and interesting products.

← Back to Blog