High-quality Krea 2 pipeline focused on extreme facial and background detail recovery, followed by a powerful SeedVR2 4K upscale.
This workflow is designed for users who want maximum sharpness and realism from Krea 2 generations without complicated setups.
What this workflow does:
- Text-to-Image generation with Krea 2 architecture
- 1st SeedVR2 pass → deblur + detail recovery
- SAM3 segmentation (Face + Hair)
- Dedicated Face Detailer (low denoise, high precision)
- Dedicated Background Detailer (separate density control)
- Final SeedVR2 4K Upscale (up to 4096px)
- Side-by-side comparison (before vs after upscale)
Key Features
- Separate density controls for Face and Background
- Clean and organized node groups
- Batch / Loop support
- Image Comparer included
- Optimized for high detail and natural skin texture
- Ready for character reference sheets or high-quality portraits
Recommended Use Cases
- Character reference sheets
- High-detail portraits
- Realistic / semi-realistic images
- Face enhancement + full scene refinement
- Final 4K delivery
Required Custom Nodes
- ComfyUI-Impact-Pack (DetailerForEach)
- ComfyUI-KJNodes (Set/Get nodes)
- rgthree-comfy
- Easy-Use (sam3, forLoop, etc.)
- SeedVR2 nodes
- ComfyUI-Image-Resize / WAS Node Suite (or equivalent)
Models needed
- Krea 2 compatible UNET
- Qwen3-VL / Lumina2 CLIP
- SeedVR2 DiT + VAE (seedvr2_ema_7b_fp16 + ema_vae_fp16)
- SAM3 model (sam3.pt)
- Optional LoRAs for style/identity
Tips
- Start with Face Density around 0.08 ~ 0.15
- Background Density around 0.12 ~ 0.25
- Keep the first SeedVR2 pass at moderate resolution for better detail recovery
- Use the Image Comparer to check the quality jump after the final 4K upscale