Open-source control LoRAs for MOSS-Image-SFT: pose, depth, canny and more
pavel_0869874 · reddit · 2026-11-10
A Reddit user released a set of control LoRAs for MOSS-Image-SFT, letting you dictate composition via a pose skeleton, depth map, or edge drawing while the prompt decides who's in the picture and what the world looks like.
Seven control types are open-sourced on Hugging Face (under thedeoxen), each repo shipping the weights plus a ready-to-use ComfyUI workflow:
- Pose
- Depth
- Canny
- Lineart
- Normal
- Scribble
- Segmentation
The demo gallery shows one pose skeleton generating wildly different images — fairies, robots, puppets — depending on the prompt.
Related event: Seven control LoRAs open-sourced for MOSS-Image-SFT(4 posts)→
More from Multimodal
- AI-generated anime pilot 'Mikawa Tsunami' shows cinematic quality — No-Past-7449 · 2026-11-10
- First look at MiniMax's H3Image: 'Stormy Weather' western-genre generation — bennash · 2026-11-10
- MiniMax H3Image early-access demo: an infinite Droste-effect spiral of Jensen Huang — bennash · 2026-11-10
- Fable 5.2 models a Bugatti Tourbillon in Blender from a single prompt in 2.5 hours — socialwithaayan · 2026-11-10
- PixVerse's PixLight 2026 AI film contest offers $300K prize pool for AI pilots and screenplays — Kavanthekid · 2026-11-10
- One prompt, one hour: Fable 5.2 generates a walkable procedural city — Merzmensch · 2026-11-10