Kromoss fine-tune of MOSS-Image-Base trends on Hugging Face
Original title: lodestones/Kromoss — new model trending #30 on Hugging Face
The 1024 px checkpoint ships two 8-step turbo LoRAs and is presented as a drop-in for MOSS-Image-SFT.
lodestones/Kromoss, trending at number 30 on Hugging Face, is a full fine-tune of MOSS-Image-Base at 1024 px. It uses the same architecture, inference pipeline, Qwen3.5-9B text encoder and FLUX.2 small VAE as MOSS-Image-SFT. The published checkpoint is update 11000 with EMA weights, trained on 704k images. The repo also includes two rank-128 8-step LoRAs distilled from MOSS-Image-Base for CFG 1 sampling.
Unconfirmed: The PartiPrompts ranking is stated by the model card, not an independent benchmark.
- The recommended LoRA is 8-steps-dmd2-lora.safetensors, further trained with a GAN stage.
- Full sampling is 50 steps at CFG 3; either LoRA uses 8 steps at CFG 1.
- An int8 ConvRot ComfyUI weight file is listed at 5.4 GB.
- The card says the DMD2 LoRA has the best PartiPrompts score.
- Vendors
- lodestones
- Organizations
- OpenMOSS-Team
From the original article
4 min read at huggingface.cotext-to-image · 5,365 downloads · 150 likes · license apache-2. 0
By lodestones · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.