Eric Qwen ControlNet Prompt Rewriter (Vision)

A ComfyUI node in Eric Qwen-Image with 11 inputs and 2 outputs.

By EricRollei·Created 5 months ago·Updated 3 months ago· 19

image

prompt
cn_type_index

◄creative_direction►

◄cn_typeSoftline►

◄api_urlhttp://localhost:1234/v1►

◄modelqwen3-vl►

◄languageEnglish►

◄temperature0.7►

◄max_tokens2048►

◄custom_instructions►

◄passthroughfalse►

◄image_max_side1024►

CategoryEric Qwen-Image

Inputs (11)

Name	Type	Default	Description
image	IMAGE		Reference image for the VLM to analyse.
creative_direction	STRING		Describe how the new image should differ from the reference. Can be brief — the VLM will expand it into a full descriptive prompt. Example: 'cyberpunk warrior in neon-lit alley' Example: 'oil painting in Rembrandt style' Example: 'same scene but underwater'
cn_type	COMBO	Softline	Which ControlNet preprocessor will be used. Informs the VLM about what structural info is preserved vs what must be described: • Softline — soft contours (idx 1) • DWPose — body skeleton only (idx 2) • Depth — spatial depth layout (idx 3) • CannyEdge — hard edges/outlines (idx 4) • PyraCanny — multi-scale edges (idx 5)
api_url	STRING	http://localhost:1234/v1	OpenAI-compatible vision API URL. Ollama: http://localhost:11434/v1 LM Studio: http://localhost:1234/v1 OpenAI: https://api.openai.com/v1
model	STRING	qwen3-vl	Vision model name. Must support image input. Ollama: qwen3-vl, qwen2.5-vl:7b, llava OpenAI: gpt-4o-mini, gpt-4o
languageopt	COMBO	English	Language for the rewritten prompt.
temperatureopt	FLOAT	0.70–2	VLM temperature. Lower = more faithful.
max_tokensopt	INT	2048256–8192	Max tokens for the VLM response.
custom_instructionsopt	STRING		Additional instructions for the VLM. Example: 'Use a cinematic photography style' Example: 'Emphasize dramatic lighting' Appended to the system prompt.
passthroughopt	BOOLEAN	false	Skip VLM rewriting. Passes creative_direction through as-is (for A/B testing).
image_max_sideopt	INT	1024256–2048	Max pixel size (longest side) for the image sent to the VLM. Larger = more detail but slower and more tokens.

Outputs (2)

Name	Type	Description
prompt	STRING	—
cn_type_index	INT	—