ComfyUI Node

Caption Creator

Creates a factual caption for each input image with a native Qwen3-VL ConvRot text encoder. Thinking is disabled and decoded text is returned unchanged.

By Merserk·Created 9 months ago·Updated 23 days ago· 5
Caption Creator
  • image
  • text
model_precisionint8
auto_downloadtrue
words100
Categoryflow-assistor/image/caption

Inputs (4)

NameTypeDefaultDescription
imageIMAGEA ComfyUI IMAGE tensor; batches are supported.
model_precisionCOMBOint8Choose the Qwen3-VL ConvRot model precision to load.
auto_downloadBOOLEANtrueDownload a missing model into models/text_encoders/flow-assistor.
wordsINT1000–200Approximate words per caption, not a hard limit. Set to 0 for an unrestricted detailed caption.

Outputs (1)

NameTypeDescription
textSTRING