ComfyUI-Qwen-Image-2.1-Prompt-Enhancer adds T2I and Edit nodes

This node provides two tools: T2I Prompt Rewrite expands short prompts into detailed descriptions, while Edit Prompt Rewrite converts vague instructions into precise prompts using source images.

The node uses Qwen3-VL 8B fine-tuned models as text encoders via ComfyUI's CLIPLoader. Models are placed in ComfyUI/models/text_encoders/.

Setup requires downloading checkpoint files from HuggingFace and installing the node. Both nodes operate within ComfyUI's built-in text encoding pipeline.

Highlights

• T2I Prompt Rewrite — expands a short text prompt into a detailed English description for text-to-image generation
• Edit Prompt Rewrite — rewrites a vague edit instruction + source image(s) into a precise, actionable prompt
• qwen3.59bqwenimage2.1pet2i.int8convrot.safetensors — 9.47 GB, text-to-image prompt rewriter
• qwen3.59bqwenimage2.1pei2i.int8convrot.safetensors — 9.47 GB, image-edit prompt rewriter
• Qwen/Qwen-Image-2.1-PE-T2I — for text-to-image prompt expansion
• Qwen/Qwen-Image-2.1-PE-I2I — for image-editing instruction rewrite
• VRAM: ~20GB for the PE model in bfloat16. A 24GB GPU is comfortable.
• VRAM is managed by ComfyUI's model management — the model is loaded and evicted like any other text encoder.

Install

From the source

Images from benjiyaya on GitHub, shown with a link to the original.