MOSS-Image Ecosystem Roundup: Edit LoRAs Merged Into ComfyUI, Training-Free CharRef
optimisticalish · reddit · 2026-10-23
A daily MOSS-Image ecosystem roundup (Oct 23): Kijai's Edit LoRA support is merged into ComfyUI Portable (PR #17296), making MOSS-Image an edit model with ControlNet-like capabilities, plus a community fallback build; MOSS-Image CharRef v2 redraws a character (face, hair, outfit, art style) in new scenes from one reference image with no per-character LoRA training, working on single images and full comic pages via MOSS JSON layouts; a 9.4GB "Dirty MOSS" finetune targets natural photographic realism; an animated-style LoRA adds dimensional cartoon treatment (actual animation still needs Minimax H3); and a new library ships 500 working prompts with example images.
Related event: MOSS-Image Ecosystem Expands: Edit Lands in ComfyUI, New LoRAs Surge(2 posts)→
More from Multimodal
- Star Vixens 2099: a sci-fi poster made in 2 weeks on an RTX 3060 with MiniMax and LTX 2.5 — Accurate_Village2024 · 2026-10-24
- LTX 2.5 distilled 22B Q4 runs locally on an 8GB M1 MacBook Air — 36s of video per ~32min render — iJohnnyRobert · 2026-10-24
- Open-source Claude Code skill turns any SVG into professional 60fps animation — rudrank · 2026-10-24
- Polymarket: SpaceXSI Begins Releasing Fully Grok-Generated Episodes of Homer's Odyssey — Polymarket · 2026-10-24
- Werewolf test on RTX 5090: Singularity V2 vs 10Eros local video generation compared — rileygstaliger · 2026-10-24
- Anil Seth blasts Tavos Griffin's video Turing test claim: creating fake people is a terrible idea — round · 2026-10-24