Nodes/Eric_Qwen_Edit_Experiments/Eric Qwen Inpaint Prompt Rewriter (Vision)
ComfyUI Node

Eric Qwen Inpaint Prompt Rewriter (Vision)

A ComfyUI node in Eric Qwen-Image with 16 inputs and 1 output.

By EricRollei·Created 5 months ago·Updated 3 months ago· 19
Eric Qwen Inpaint Prompt Rewriter (Vision)
  • image
  • mask
  • prompt
change_description
api_urlhttp://localhost:1234/v1
model
outpaint_description
languageEnglish
temperature0.7
max_tokens2048
outline_color_r255
outline_color_g50
outline_color_b50
outline_width3
custom_instructions
passthroughfalse
image_max_side1024
CategoryEric Qwen-Image

Inputs (16)

NameTypeDefaultDescription
imageIMAGESource image for the VLM to analyse.
change_descriptionSTRINGDescribe what you want to change in the masked area. Example: 'Replace the car with a green taxi' Example: 'Change the sky to a dramatic sunset'
api_urlSTRINGhttp://localhost:1234/v1OpenAI-compatible vision API URL. LM Studio: http://localhost:1234/v1 Ollama: http://localhost:11434/v1 OpenAI: https://api.openai.com/v1
modeloptSTRINGModel name override. Leave empty to auto-detect the loaded model from the API server. Ollama: qwen3-vl, qwen2.5-vl:7b OpenAI: gpt-4o-mini, gpt-4o
maskoptMASKInpaint mask (white = area to change). A red outline will be drawn on the image for the VLM. Not needed for outpainting.
outpaint_descriptionoptSTRINGWhat to add in the outpaint expansion area. Example: 'Mountain landscape extending into the distance' Only used when no mask is provided or when you want outpainting context in the prompt.
languageoptCOMBOEnglishLanguage for the rewritten prompt.
temperatureoptFLOAT0.70–2VLM temperature. Lower = more faithful.
max_tokensoptINT2048256–8192Max tokens for the VLM response.
outline_color_roptINT2550–255Mask outline red channel.
outline_color_goptINT500–255Mask outline green channel.
outline_color_boptINT500–255Mask outline blue channel.
outline_widthoptINT31–10Pixel width of the mask outline.
custom_instructionsoptSTRINGAdditional instructions for the VLM. Example: 'Use a cinematic photography style' Appended to the system prompt.
passthroughoptBOOLEANfalseSkip VLM rewriting. Passes change_description through as-is (for A/B testing).
image_max_sideoptINT1024256–2048Max pixel size (longest side) for the image sent to the VLM. Larger = more detail but slower and more tokens.

Outputs (1)

NameTypeDescription
promptSTRING