Hugging Face's logo

MATLOWAI
/
MOSS-Image-SFT-DeBand

Text-to-Image
Diffusion Single File
comfyui
lora
moss-image
Model card Files
xet
 Community
matlod commited on
Commit
293dd2e
·
verified ·
1 Parent(s): 17bb919

README: direct download links for every file the workflows need

Browse files
Files changed (1) hide show
  1. README.md +20 -15
README.md CHANGED
@@ -62,17 +62,17 @@ The speed presets keep SFT's composition fairly well; the Turbo presets are fast
62
 
63
  ### `moss_banding_fix_presets.json`: stock ComfyUI, no custom nodes
64
 
65
- Put these files in your ComfyUI `models` folder:
66
 
67
- | File | Folder | Get it from |
68
  |---|---|---|
69
- | `moss_image_sft_bf16.safetensors` | `diffusion_models` | [Comfy-Org/MOSS-Image](https://huggingface.co/Comfy-Org/MOSS-Image) |
70
- | `qwen3.5_9b_int8_convrot.safetensors` | `text_encoders` | [Comfy-Org/MOSS-Image](https://huggingface.co/Comfy-Org/MOSS-Image) |
71
- | `flux2-vae.safetensors` | `vae` | [Comfy-Org/MOSS-Image](https://huggingface.co/Comfy-Org/MOSS-Image) |
72
- | `moss_image_sft_banding_fix_L37-41_bf16.safetensors` | `loras` | this repo |
73
- | `moss_image_sft_banding_fix_L32-36_extra_bf16.safetensors` | `loras` | this repo |
74
- | `lodestones_moss_sft_8step_dmd_lora_latest.safetensors` | `loras` | `lora_latest.safetensors` from [lodestones/MOSS-Image-SFT-8-steps-dmd](https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd), **renamed** |
75
- | `darask0_moss_sft_4step_dmad_lora.safetensors` | `loras` | `lora.safetensors` from [darask0/MOSS-Image_4-step_adapter_lora](https://huggingface.co/darask0/MOSS-Image_4-step_adapter_lora), **renamed** |
76
 
77
  The two speed LoRAs are both called `lora.safetensors` upstream, so rename them as shown. A LoRA at strength 0 is skipped, but ComfyUI still checks that every file exists. If you don't want a speed LoRA, delete its node and don't pick its preset.
78
 
@@ -80,12 +80,17 @@ The two speed LoRAs are both called `lora.safetensors` upstream, so rename them
80
 
81
  Everything above, plus:
82
 
83
- - **Custom nodes:** [OpenMOSS's `moss-image` nodes](https://github.com/OpenMOSS/MOSS-Image/tree/main/comfyui/moss-image) and darask0's [`rapid-moss` node](https://huggingface.co/darask0/MOSS-Image_4-step_adapter_lora/tree/main/comfyui/rapid-moss), each copied into `ComfyUI/custom_nodes/`.
84
- - **`split_qkv/moss_image_sft_bf16_split_qkv.safetensors`** (this repo) → `models/diffusion_models`. The rapid-moss adapters patch q, k and v separately, so they need the SFT model with q/k/v stored apart. This file is the official [OpenMOSS-Team/MOSS-Image-SFT](https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT) transformer as a single file, with the same 463 tensors. The Comfy-Org file fuses q/k/v; with it, the rapid-moss adapters produce noise or a black image without raising an error.
85
- - **`moss_image_sft_banding_fix_L37-41_bf16_splitqkv.safetensors`** (this repo) → `models/loras`. The same banding fix, laid out for the split model.
86
- - **Adapter folders**, each holding `lora.safetensors`, `time_embedding.safetensors` and `dataspace.safetensors`:
87
- - `models/rapid_moss/lodestones-8step-latest/`: from lodestones' repo, with `lora_latest.safetensors` saved as `lora.safetensors`.
88
- - `models/rapid_moss/rapid-moss-4step/`: from darask0's repo.
 
 
 
 
 
89
 
90
  Only the branch for the chosen preset runs, so the normal presets don't load the split model.
91
 
 
62
 
63
  ### `moss_banding_fix_presets.json`: stock ComfyUI, no custom nodes
64
 
65
+ Put these files in your ComfyUI `models` folder (each link downloads the file directly):
66
 
67
+ | Save as | Into folder | Download |
68
  |---|---|---|
69
+ | `moss_image_sft_bf16.safetensors` | `diffusion_models` | [Comfy-Org](https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/diffusion_models/moss_image_sft_bf16.safetensors) |
70
+ | `qwen3.5_9b_int8_convrot.safetensors` | `text_encoders` | [Comfy-Org](https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/text_encoders/qwen3.5_9b_int8_convrot.safetensors) |
71
+ | `flux2-vae.safetensors` | `vae` | [Comfy-Org](https://huggingface.co/Comfy-Org/MOSS-Image/resolve/main/vae/flux2-vae.safetensors) |
72
+ | `moss_image_sft_banding_fix_L37-41_bf16.safetensors` | `loras` | [this repo](https://huggingface.co/MATLOWAI/MOSS-Image-SFT-banding-experimental/resolve/main/moss_image_sft_banding_fix_L37-41_bf16.safetensors) |
73
+ | `moss_image_sft_banding_fix_L32-36_extra_bf16.safetensors` | `loras` | [this repo](https://huggingface.co/MATLOWAI/MOSS-Image-SFT-banding-experimental/resolve/main/moss_image_sft_banding_fix_L32-36_extra_bf16.safetensors) |
74
+ | `lodestones_moss_sft_8step_dmd_lora_latest.safetensors` | `loras` | [`lora_latest.safetensors` from lodestones](https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/resolve/main/lora_latest.safetensors), **renamed** |
75
+ | `darask0_moss_sft_4step_dmad_lora.safetensors` | `loras` | [`lora.safetensors` from darask0](https://huggingface.co/darask0/MOSS-Image_4-step_adapter_lora/resolve/main/lora.safetensors), **renamed** |
76
 
77
  The two speed LoRAs are both called `lora.safetensors` upstream, so rename them as shown. A LoRA at strength 0 is skipped, but ComfyUI still checks that every file exists. If you don't want a speed LoRA, delete its node and don't pick its preset.
78
 
 
80
 
81
  Everything above, plus:
82
 
83
+ - **Custom nodes**, each folder copied into `ComfyUI/custom_nodes/`:
84
+ - OpenMOSS's [`moss-image`](https://github.com/OpenMOSS/MOSS-Image/tree/main/comfyui/moss-image) (the `comfyui/moss-image` folder of their GitHub repo);
85
+ - darask0's [`rapid-moss`](https://huggingface.co/darask0/MOSS-Image_4-step_adapter_lora/tree/main/comfyui/rapid-moss) (the `comfyui/rapid-moss` folder of their Hugging Face repo).
86
+ - **[`moss_image_sft_bf16_split_qkv.safetensors`](https://huggingface.co/MATLOWAI/MOSS-Image-SFT-banding-experimental/resolve/main/split_qkv/moss_image_sft_bf16_split_qkv.safetensors)** (10.2 GB, this repo) → `models/diffusion_models`. The rapid-moss adapters patch q, k and v separately, so they need the SFT model with q/k/v stored apart. This file is the official [OpenMOSS-Team/MOSS-Image-SFT](https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT) transformer as a single file, with the same 463 tensors. The Comfy-Org file fuses q/k/v; with it, the rapid-moss adapters produce noise or a black image without raising an error.
87
+ - **[`moss_image_sft_banding_fix_L37-41_bf16_splitqkv.safetensors`](https://huggingface.co/MATLOWAI/MOSS-Image-SFT-banding-experimental/resolve/main/moss_image_sft_banding_fix_L37-41_bf16_splitqkv.safetensors)** (this repo) → `models/loras`. The same banding fix, laid out for the split model.
88
+ - **Adapter folders**, each holding three files with exactly these names:
89
+
90
+ | Folder | `lora.safetensors` | `time_embedding.safetensors` | `dataspace.safetensors` |
91
+ |---|---|---|---|
92
+ | `models/rapid_moss/lodestones-8step-latest/` | [lodestones' `lora_latest.safetensors`](https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/resolve/main/lora_latest.safetensors), **renamed** to `lora.safetensors` | [download](https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/resolve/main/time_embedding.safetensors) | [download](https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/resolve/main/dataspace.safetensors) |
93
+ | `models/rapid_moss/rapid-moss-4step/` | [darask0's `lora.safetensors`](https://huggingface.co/darask0/MOSS-Image_4-step_adapter_lora/resolve/main/lora.safetensors) | [download](https://huggingface.co/darask0/MOSS-Image_4-step_adapter_lora/resolve/main/time_embedding.safetensors) | [download](https://huggingface.co/darask0/MOSS-Image_4-step_adapter_lora/resolve/main/dataspace.safetensors) |
94
 
95
  Only the branch for the chosen preset runs, so the normal presets don't load the split model.
96