Model card: ComfyUI section
Browse files
README.md
CHANGED
|
@@ -70,6 +70,19 @@ python inference.py --model-dir models/MiniMax-H3 --lora ckpt/minimax_h3/dmad_mi
|
|
| 70 |
--prompt-file prompts/dmad_sweater.txt --seed 42 --output-dir outputs/dmad_sweater
|
| 71 |
```
|
| 72 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
## Wan2.1 (text-to-video)
|
| 74 |
|
| 75 |
Full generators (EMA, spectral norm folded in) in the `.pth` format of
|
|
|
|
| 70 |
--prompt-file prompts/dmad_sweater.txt --seed 42 --output-dir outputs/dmad_sweater
|
| 71 |
```
|
| 72 |
|
| 73 |
+
## ComfyUI
|
| 74 |
+
|
| 75 |
+
`comfyui/dmad_minimax_h3_4step_{lora_critic,full_critic}_comfyui.safetensors` are the same two LoRAs converted exactly
|
| 76 |
+
to ComfyUI's MiniMax-H3 key layout (rank 128; q/k/v fused into `attn.qkv_proj` adapters of rank 384, alpha = rank,
|
| 77 |
+
scale 1.0 — no rank reduction; 2.0 GB each). Load with `LoraLoaderModelOnly` at strength 1.0, cfg 1.0,
|
| 78 |
+
`ModelSamplingMiniMaxH3` with shift 12 / audio shift 2, and sample with the **DMAD Sampler** + **DMAD Sigmas** nodes
|
| 79 |
+
from [`comfyui/ComfyUI-DMAD`](https://github.com/Yzmblog/DMAD/tree/main/comfyui/ComfyUI-DMAD) (the re-noise multistep
|
| 80 |
+
rule the students were trained with; stock samplers are not their operating point).
|
| 81 |
+
|
| 82 |
+
```bash
|
| 83 |
+
hf download ZhengmingYu/DMAD --include "comfyui/*" --local-dir /path/to/ComfyUI/models/loras
|
| 84 |
+
```
|
| 85 |
+
|
| 86 |
## Wan2.1 (text-to-video)
|
| 87 |
|
| 88 |
Full generators (EMA, spectral norm folded in) in the `.pth` format of
|