Nodes/WAS Node Suite v3/Generate Edit Instruction
ComfyUI Node Runs on cloud

Generate Edit Instruction

Turn a short edit request into an explicit instruction an image editing model follows fully, written by a language model that reads the picture: what changes, implied effects included, and what stays. With a judge's notes wired in, it corrects the last attempt.

By WASasquatch·Created 4 years ago·Updated 3 days ago· 1,900
Generate Edit Instruction
  • clip
  • image
  • instruction
◄instruction►
◄temperature0.00►
◄seed0►
◄max_length256►
◄feedback—►
◄last_instruction—►
CategoryWAS Suite/Text

Inputs (8)

NameTypeDefaultDescription
clipCLIPqwen3.5_9b_int8_convrot.safetensors from Load CLIP with type krea2, or another Qwen-VL model.
imageIMAGEThe picture to edit. The first of a batch is read.
instructionSTRINGThe edit as written, as 'Make it a snowy winter night'.
temperatureFLOAT0.000–20 = the likeliest words, the same instruction every time; 0.7 = varied wording per seed.
seedINT00–18446744073709550000The seed sampling draws from. Ignored at temperature 0.
max_lengthINT25632–2048The most tokens the instruction may run to: 256 covers about 180 words.
feedbackoptSTRINGJudge Images' notes on the last attempt, as 'the sky is blown out and shows no snow'. Unwired or empty = a first attempt.
last_instructionoptSTRINGThe instruction the last attempt was made from, as 'Add falling snow and a dark blue sky'.

Outputs (1)

NameTypeDescription
instructionSTRINGThe explicit instruction, for the edit model's prompt and Judge Images.