ComfyUI Node
Sampler_DisPose
A ComfyUI node in DisPose with 12 inputs and 3 outputs.
Sampler_DisPose
- pipe
- ref_video
- ref_image
- latent
- ref_image
- poses
◄resolution576►
◄seed42►
◄tile_size16►
◄tile_overlap6►
◄num_inference_steps25►
◄noise_aug_strength0.00►
◄min_guidance_scale2.0►
◄max_guidance_scale2.0►
◄cpu_offloadmodel►
CategoryDisPose
Inputs (12)
| Name | Type | Default | Description |
|---|---|---|---|
| pipe | DisPosePipeline | output from (Down)Loader_DisPose | |
| ref_video | IMAGE | video or concatenated images as reference poses, recommend to use VideoHelperSuite | |
| ref_image | IMAGE | one image as driving image, keeping everything but pose | |
| resolution | INT | 576256–640 | shorter edge pixels, aspect ratio is 16/9 |
| seed | INT | 42 | set control_after_generation as fixed to make result reproducible |
| tile_size | INT | 161–640 | larger size makes better motion quality but requires more vram |
| tile_overlap | INT | 60–64 | larger overlap makes better motion quality but increase processing time, vram remains the same |
| num_inference_steps | INT | 251–100 | — |
| noise_aug_strength | FLOAT | 0.000–1 | larger value makes better motion quality but less alike to ref_image |
| min_guidance_scale | FLOAT | 2.00.1–10 | — |
| max_guidance_scale | FLOAT | 2.00.1–10 | — |
| cpu_offload | COMBO | model | sequence is slowest but consume the least VRAM, model also spare some vram and slightly slower than none |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| latent | LATENT | — |
| ref_image | IMAGE | — |
| poses | IMAGE | — |