ComfyUI Node
MS-Diffusion Conditioning (Native)
Build MS-Diffusion image prompt tokens from CLIP-Vision penultimate hidden states + CLIP-L phrase pooled embeddings + layout boxes. Connect prompt_clean to your positive CLIPTextEncode so the two match.
MS-Diffusion Conditioning (Native)
- ms_adapter
- clip
- clip_vision_output
- ms_embeds
- prompt_clean
◄promptbest quality, high quality, a [dog] on the beach►
◄boxes0.25, 0.25, 0.75, 0.75►
◄mask_threshold0.50►
◄drop_grounding_tokensfalse►
CategoryMSdiffusion/Native
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| ms_adapter | MS_ADAPTER | — | |
| clip | CLIP | — | |
| clip_vision_output | CLIP_VISION_OUTPUT | — | |
| prompt | STRING | best quality, high quality, a [dog] on the beach | — |
| boxes | STRING | 0.25, 0.25, 0.75, 0.75 | one 'x1,y1,x2,y2' per subject, separated by ';'. Ignored when mask_threshold is 0. |
| mask_threshold | FLOAT | 0.500–1 | 0 disables layout guidance entirely (upstream behaviour). >0 enables box guidance + pseudo attention mask. |
| drop_grounding_tokens | BOOLEAN | false | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| ms_embeds | MS_EMBEDS | — |
| prompt_clean | STRING | — |