Stops reading your prompt this many layers before the end of the text encoder.
Typical 1-2Default 1
The encoder builds meaning up in layers, and stopping early hands the model a looser, less literal reading of what you asked for. 1 is standard and is what most models were trained against. Many anime-derived checkpoints were trained at 2, and on those the difference is not subtle. Where a model states a value, the panel marks it on the CLIP skip control as a suggestion.
Sampling · Interactive explorer CLIP skip
CLIP skip 2 512 × 512
1Drag to compare3
Skips the last layer. On an anime-lineage model this is the intended setting; elsewhere it just loosens the prompt.