🎨 Style Transfer Prompter
Generate style-only prompts from a reference image for FLUX, Krea, and other Image-to-Image models.
Unlike traditional image captioning, Style Transfer Prompter intentionally ignores the image content. Instead, it instructs a local VLM (Qwen3-VL) to describe only the visual style—brushwork, line quality, edge treatment, rendering technique, textures, and color palette.
As a result, the target image keeps its original subject while adopting the artistic style of the reference image.
✨ Features
✅ Style-only prompt generation
No characters, objects, poses, or scene descriptions—only stylistic attributes.
✅ Four analysis modes
🎨 Artistic / Illustration
🧊 3D / Render / CGI
📷 Photorealistic / Cinematic
✏️ Anime / Manga
✅ Identity preservation
keep_identity helps preserve the identity of the target subject.
✅ Painterly edge detection
painterly_edges describes unfinished brush strokes, watercolor bleeding, rough edges, and loose paint transitions.
✅ Color control
keep_original_colors
preserve the original colors of the target image
or adopt the palette of the reference image
✅ Custom prompt extension
extra_prompt appends your own instructions to the generated prompt.
📦 Dependencies
This node requires:
ComfyUI-QwenVL — local VLM for style analysis
(https://github.com/1038lab/ComfyUI-QwenVL)
ZImageResolutionScaler — automatic resolution matching
(https://civarchive.com/models/2344436/z-image-resolution-picker)
FLUX 2 Klein 9B Distilled — Image Editing model
⚙️ Installation
Download the archive.
Extract it.
Copy ComfyUI-StyleTransferPrompter into:
ComfyUI/custom_nodes/
Restart ComfyUI.
Make sure ComfyUI-QwenVL is installed.
Download a compatible Qwen3-VL model.
🖼 Included Workflow
The included style_transfer_workflow.json is based on the standard FLUX 2 Klein Image Editing workflow.
Pipeline
Style Reference Image
│
▼
AILab_QwenVL_Advanced
│
▼
Style Transfer Prompter
│
▼
Generated Style Prompt
│
▼
FLUX
▲
│
Target Image
│
▼
ZImageResolutionScaler
🧩 Workflow Nodes
Node Purpose
Style Reference Image Reference image for style analysis
Target Image + Mask Image to be stylized
AILab_QwenVL_Advanced Generates a style description using a local VLM
Style Transfer Prompter Converts the description into a FLUX-compatible style prompt
ZImageResolutionScaler Automatically matches processing resolution
FLUX 2 Klein 9B Performs Image-to-Image stylization
⚙️ Node Parameters
🎨 style_mode
Selects the category used for style analysis.
👤 keep_identity
Helps preserve the identity of the target subject.
🖌 painterly_edges
Emphasizes unfinished edges, brush strokes, watercolor bleeding, and loose paint transitions.
🌈 keep_original_colors
Keeps the original colors of the target image instead of adopting the reference palette.
✍️ extra_prompt
Appends additional custom instructions to the generated prompt.
🎭 Mask Usage
Use Target Image and Mask (standard Load Image node) and create a mask via Open in MaskEditor.
When connected to VAE Encode, only the masked region is stylized, while the rest of the image remains unchanged.
