Method: stars today from the daily GitHub Trending dump issues Rain1er/trending #400 (“2026年10月9日”, updated 2026-10-09 17:22 UTC) and happydog-intj/ai-xiaohongshu-daily #548 (“2026-10-10”), compared with the previous day’s issue and our streak log. Repos with 40k+ total stars, long-running chart leaders, non-AI repos and repos with only modest gains are excluded. Excluded today: morluto/rea (+15,335, but now about 41k total stars, so the 40k rule applies), boykopovar/AnyPS5 (+5,925, non-AI PS5 porting tool), mattpocock/skills, diagram-design, open-code-review, agent-skills and litellm (all 40k+), anthropics/knowledge-work-plugins (+714, up from +309, but only 2.5% of its total) and Robbyant/lingbot-map (+109).
storytold/artcraft — +3,723 ⭐ today
URL: https://github.com/storytold/artcraft
Total stars: 10,617 · vs. previous day: not on the previous day’s list (new entry; +35% of its total in one day) · Language: Rust · License: ArtCraft License (WIP, a custom “fair source” license, not OSI open source)
ArtCraft bills itself as “the IDE for interactive AI image and video creation,” a desktop creative tool that gives artists, designers and filmmakers control over composition and character placement that text prompts alone struggle to deliver. Users arrange images, cutouts, simple 3D meshes and posed characters into a 2D/3D scene (scene blocking), then generate or edit images and video from that layout. The repository is a Rust workspace (crates) with a separate frontend, and generation itself is handled by external models chosen from one interface — Nano Banana Pro, GPT-Image-1.5, Seedream, Flux, Veo, Kling, Sora 2, Grok Imagine, Midjourney and WorldLabs Marble (Gaussian-splat worlds) — making it a multi-model front end. Key features:
- Image to Location (placing characters in consistent environments) and 2D/3D image compositing
- Image to 3D Mesh, character posing and character identity transfer
- Scene blocking with kit bashing, background removal and mixed-asset crafting
- Text to Image across a dozen-plus models, inpainting/drawing/masking edits, and Image to Video
What sets it apart is wrapping many commercial generators in a single scene editor, so the workflow is about directing a shot rather than rerolling prompts. With Windows and macOS builds available and support for the latest video models such as Sora 2, Veo and Kling, it drew a wave of creators and picked up more than 3,700 stars in a day (its best Trendshift result was #11 Rust repo of the day on 10-06, with no #1 badge).
Practical takeaway: ad and content teams can trial it as a previs tool to quickly compare storyboard-faithful, character-consistent drafts across several generative models (check the custom license before commercial use).
Tags: #AIImageGeneration #AIVideo #SceneBlocking #CharacterConsistency #CreativeTools #Rust
Leave a Reply