ComfyUI Node Runs on cloud
MOSS-Image Prompt Weight
Per-token prompt weighting for MOSS-Image via attention value scaling, same approach as Krea2 Prompt Weight. Use (word:-1) to remove a concept, (word:1.5) to emphasize one -- works through the Qwen3.5 encoder where normal weighting doesn't. Outputs the patched model + conditioning; wire both to the sampler.
MOSS-Image Prompt Weight
- clip
- model
- model
- conditioning
◄text►
◄strength1.00►
CategoryKJNodes/experimental
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | — | |
| model | MODEL | — | |
| text | STRING | Prompt with per-token weights in parentheses, e.g. (word:-1) removes/represses a concept, (word:2) emphasizes it, plain text = 1.0. Works on MOSS-Image where ComfyUI's normal (word:weight) does nothing (the Qwen3.5 LLM encoder ignores it). weight<1 scales the token's attention VALUE (subtracts/removes at <0); weight>1 boosts how much the image ATTENDS to the token (adds more of it). Only positive conds are weighted, negative prompts pass through untouched. | |
| strength | FLOAT | 1.000–4 | Global multiplier on the weighting effect. Effect compounds over all 42 blocks; lower if results break up, raise for a stronger effect. Removal (weight<0) is the reliable direction; emphasis (weight>1) works but is looser. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| conditioning | CONDITIONING | — |