Kromoss

Kromoss is a full fine-tune of MOSS-Image-Base at 1024 px. It uses the same architecture, inference pipeline, text encoder (Qwen3.5-9B) and VAE (FLUX.2 small decoder) as MOSS-Image-SFT, and this repo has the same layout, so it is a drop-in replacement. The fine-tune mainly fixes the Base model's blurriness.

The repo also ships 8-steps-dmd-lora.safetensors, an 8-step turbo LoRA (data-space DMD distillation of MOSS-Image-Base): 8 steps, CFG 1, no negative branch.

Checkpoint: update 7000 of the fine-tune (EMA weights), 448k training images.

Quick start

pip install "diffusers>=0.38" "transformers>=5.3" "accelerate>=1.10" "safetensors>=0.6" "Pillow>=10"

import torch
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained(
    "lodestones/Kromoss",
    custom_pipeline="pipeline_moss_image",
    trust_remote_code=True,
    torch_dtype={"default": torch.bfloat16, "vae": torch.float32},
)
pipe.scheduler = type(pipe.scheduler).from_config(pipe.scheduler.config)
pipe.to("cuda")

image = pipe(
    "a tabby cat holding a sign that says KROMOSS",
    height=1024,
    width=1024,
    num_inference_steps=50,
    guidance_scale=3.0,
    uncond_skip_layers=[22],
    generator=torch.Generator("cpu").manual_seed(42),
).images[0]
image.save("kromoss.png")

The transformer and text encoder use bfloat16; the VAE and initial noise use float32.

ComfyUI

Native ComfyUI support for MOSS-Image is in current ComfyUI master. The single files use the same layout as Comfy-Org/MOSS-Image: fused proj.qkv, bf16.

kromoss_bf16.safetensors drops in for moss_image_sft_bf16.safetensors in the Comfy-Org workflows (50 steps, CFG 3). With the LoRA, use KSampler with euler / simple, 8 steps, CFG 1. ComfyUI maps the LoRA's separate q/k/v onto the fused qkv slices.

convert_to_comfy.py converts any MOSS-Image diffusers transformer/ folder (e.g. your own fine-tune) into this single-file format (needs only torch and safetensors):

python convert_to_comfy.py path/to/transformer my_model_bf16.safetensors

On MOSS-Image-SFT's transformer/ it reproduces Comfy-Org's moss_image_sft_bf16.safetensors byte for byte; --compare FILE checks keys, shapes and dtypes against a reference file.

kromoss_int8_convrot.safetensors is ConvRot W8A8, the format of Comfy-Org's moss_image_sft_int8_convrot.safetensors. The 168 block linears (proj.{qkv,o,ffn_up_gate,ffn_down}) are stored as int8 with per-row scales, after a 256-wide regular Hadamard rotation of the input dimension. ComfyUI rebuilds the rotation at load time and applies it to the activations online, so no rotation matrices are stored. Each block also selects ComfyUI's int8 attention. It drops in for the int8 Comfy-Org file and works with the LoRA. quantize_convrot.py makes it from a bf16 single file; on moss_image_sft_bf16.safetensors it reproduces Comfy-Org's int8 file byte for byte:

python quantize_convrot.py my_model_bf16.safetensors my_model_int8_convrot.safetensors

8-step turbo LoRA

8-steps-dmd-lora.safetensors: fp32, rank 128 / alpha 128 on every block's proj.{q,k,v,o,ffn_up_gate,ffn_down}, kohya keys (diffusion_model.<module>.lora_down.weight / lora_up.weight / alpha). In ComfyUI, load it with a LoRA loader at strength 1. With diffusers, apply it unmerged. Do not merge it into the bf16 weights: the delta is close to bf16's rounding step.

from safetensors.torch import load_file

def apply_lora(transformer, path, strength=1.0):
    state = load_file(path, device="cuda")
    modules = dict(transformer.named_modules())
    for key in [k for k in state if k.endswith(".lora_down.weight")]:
        name = key[len("diffusion_model."):-len(".lora_down.weight")]
        down, up = state[key].float(), state[key.replace("lora_down", "lora_up")].float()
        scale = strength * state[f"diffusion_model.{name}.alpha"].item() / down.shape[0]
        def hook(module, args, out, down=down, up=up, scale=scale):
            return out + (scale * (args[0].float() @ down.T) @ up.T).to(out.dtype)
        modules[name].register_forward_hook(hook)

apply_lora(pipe.transformer, "8-steps-dmd-lora.safetensors")   # hf_hub_download("lodestones/Kromoss", ...)
image = pipe(prompt, height=1024, width=1024, num_inference_steps=8, guidance_scale=1.0,
             generator=torch.Generator("cpu").manual_seed(42)).images[0]

Sampling settings

PartiPrompts (100 prompts, CLIP ViT-L/14 score; Kromoss rows measured at update 2000)

Columns in the sheet below follow the table's order.

Training

• Full fine-tune of every transformer parameter, 1024 px (images under 896² dropped).
• The MOSS-Image report's final-stage recipe, scaled to 64 images per update:
  • hybrid Muon-AdamW, lr 9e-6, warmup 200;
  • truncated logit-normal timesteps;
  • final-block q/k-norm decay to one;
  • EMA 0.999 (the published weights are the EMA).

License and acknowledgements

The weights and the bundled inference code (pipeline_moss_image.py, from MOSS-Image-SFT) are released under the Apache License, Version 2.0. Built on MOSS-Image by the OpenMOSS Team.