Pipeline Stages

  1. Load Content Image: the image whose structure you want to keep
  2. Load Style Reference: the artistic image whose style you want to adopt
  3. Preprocess ControlNet: run canny or depth on the content image (control strength 0.6-0.8)
  4. Load Checkpoint: SDXL or SD1.5 base model
  5. Positive Prompt: describe the style reference (e.g., "oil painting, impasto brushstrokes, warm palette")
  6. Negative Prompt: "blurry, deformed, low quality, watermark"
  7. VAE Encode (content): encode content image to latent
  8. KSampler: denoise 0.55-0.7, steps 25-30, CFG 7-9, with ControlNet applied
  9. VAE Decode: latent to image
  10. Upscale (optional): 4x-UltraSharp for final detail
  11. Save Image: output to disk

Recommended Parameters

  • ControlNet strength: 0.7 (structure preservation)
  • Denoisestrength: 0.6
  • Steps: 28
  • CFG: 8

FAQ

What is a ComfyUI style transfer pipeline?

A multi-stage ComfyUI workflow that takes a content image and a style reference, then generates a new image preserving the content's structure while adopting the reference's artistic style. This template chains img2img, ControlNet, and upscaling nodes.

Which ControlNet works best for style transfer?

For preserving structure, use ControlNet with canny or depth preprocessor on the content image. Set control strength around 0.6-0.8 to keep composition while allowing style to dominate.

All Workflows