Radar / AI media tools / Kromoss

Kromoss: community MOSS-Image fine-tune with 8-step turbo LoRAs

Community 5B text-to-image model from 'lodestones', trending on Hugging Face Oct 21: a full fine-tune of MOSS-Image-Base at 1024px with the same architecture and pipeline as MOSS-Image-SFT, so it is a drop-in replacement. Ships two 8-step turbo LoRAs distilled with DMD/DMD2 (8 steps, CFG 1) and native ComfyUI support with int8 ConvRot single files. PartiPrompts CLIP: 28.13 for Kromoss + DMD2 LoRA vs 27.19 for the base. Weights and inference code under Apache 2.0.

new launchmedia · imageJEV traction 0.51added 2026-10-21

Open Kromoss →View on the radar

Why it matters

A community fine-tune that actually ships the fast path: DMD-distilled 8-step turbo LoRAs plus byte-exact ComfyUI single files, so the MOSS-Image stack runs in 8 steps instead of 50 with slightly better PartiPrompts scores.

What you could build with it

An indie developer building a sticker/icon generator could drop Kromoss into the ComfyUI workflows already built for MOSS-Image-SFT and compare 8-step outputs against the base model on text-rendering accuracy.

Does it hold up?

Usable now — it is a drop-in replacement for MOSS-Image-SFT with documented ComfyUI workflows, but all quality numbers are the author's own PartiPrompts measurements; no independent community bake-offs yet.

Built with Kromoss

lodestones/Kromoss (Hugging Face model card)huggingface · The model card itself: full fine-tune recipe (Muon-AdamW, 704k images, EMA), 8-step turbo LoRAs, native ComfyUI support and PartiPrompts numbers.
OpenMOSS-Team/MOSS-Image-SFTgithub · The upstream OpenMOSS-Team SFT checkpoint this repo fine-tunes from — same pipeline, drop-in layout.
Comfy-Org/MOSS-Imagegithub · Official ComfyUI packaging of the MOSS-Image family; Kromoss single files drop into these workflows unchanged.

Learn more

model card (Hugging Face) →

First spotted on hf: source.

More AI media tools

VoiceStudio — open-source, fully-local ElevenLabs alternativeOpen-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation…media · JEV 0.73OpenMontage: open-source agentic video production systemTurns an AI coding assistant into a full video studio: 12 pipelines and 100+ tools for scripting, footage, narration…media · JEV 0.73LingBot-Map: streaming 3D reconstruction foundation model (ECCV 2026 best-paper candidate)Feed-forward 3D foundation model that reconstructs scenes from streaming video at ~20 FPS with drift correction over…media · JEV 0.7HyperFramesHeyGen's open-source framework that turns HTML/CSS/GSAP into deterministic MP4 video via 21 agent skills and a router…media · JEV 0.69ArtCraft pivots to seven open-source Adobe clones built with Claude Opus 5.5Brandon Thomas pivoted his AI art IDE into a seven-app open-source creative suite…media · JEV 0.67SolidCraft: open-source parametric CAD in Rust that agents can drive over MCPSolidCraft is an open-source parametric 3D CAD application written in pure Rust (desktop, browser via WebAssembly…media · JEV 0.67

Get the week's best AI launches, plus 3 ideas worth building

One email every Saturday. Ranked by traction, not hype. Free.