Flux TTP Tile Upscale | Face Restoration
The Flux Upscaler with TTP Tile technology solves the common problem of distorted faces in AI-generated images. This specialized workflow combines tile-based processing with Flux's powerful image enhancement to repair facial features while upscaling resolution to 4K. The process preserves overall composition while selectively redrawing problematic areas, making it ideal for fixing small, distorted faces in otherwise good compositions. Compatible with various model bases for different facial styles.ComfyUI Face Restore Workflow

- Fully operational workflows
- No missing nodes or models
- No manual setups required
- Features stunning visuals
ComfyUI Face Restore Examples







ComfyUI Face Restore Description
ComfyUI Flux-TTP-Upscale | Advanced Face Restore & 4K Image Enhancement
1. What is the ComfyUI Flux-TTP-Upscale Face Restore Workflow?
The Flux-TTP-Upscale workflow offers an advanced Face Restore pipeline within the ComfyUI environment. It integrates Flux's face restoration technology with TTP (Tile-to-Patch) enhancement to fix distorted or low-quality faces in AI-generated images. This is especially effective for group portraits, profile shots, or any visuals with facial artifacts.
By combining FluxGuidance, tile-aware image enhancement, and LoRA-based identity control, Flux-TTP-Upscale Face Restore delivers reliable Face Restore performance while upscaling to crisp 4K resolution.
2. Key Face Restore Features of ComfyUI Flux-TTP-Upscale
- High-Precision Face Restore: Detects and restores small or distorted faces without harming overall image composition.
- 4K Image Upscaling: Enhances resolution through TTP tile workflows and super-resolution models.
- Tile-Based Patch Enhancement: Splits the image into tiles to reduce artifacting, ensuring local Face Restore improvements blend seamlessly.
- LoRA Switching for Identity Preservation: Select the right LoRA models for Asian or non-Asian faces to improve Face Restore accuracy across different ethnicities.
3. Getting Started with the Face Restore Workflow
IMPORTANT NOTE: This Face Restore workflow handles both image enhancement and face repair simultaneously. Proper input and model selection ensure optimal results.
Quick Start Guide:
- Upload Image for Face Restore:
Use the
Load Image
node to input a low-resolution portrait, group photo, or any AI-generated image needing facial repair. - Choose the Correct LoRA Model:
- Use flux1-dev-fp8 for restoring Asian faces.
- Use original flux for general or non-Asian faces.
- Preprocessing Settings (Optional): Images are automatically resized to 1024x1024 and scaled to an 8MP target for better Face Restore quality.
- Run the Face Restore Pipeline:
Click
Queue Prompt
to initiate the restoration and upscale process. - Save Your Output:
Restored images are saved via the
Save Image
node.
4. Node Reference & Parameters for Face Restore
Guidance and Denoising
FluxGuidance
: Drives facial restoration accuracy during generation.BasicGuider
: Adds global image consistency around the restored face.SamplerCustomAdvanced
: Useseuler
sampler with fine-tuned denoise strength (denoise = 0.3
).
Preprocessing for Better Face Restore
Resize Image
: Sets up correct image dimensions for effective tile repair.Upscale Model
: Uses4xNMKD-Superscale
to refine face patches.Scale to Total Pixels
: Ensures final resolution is high enough for detailed Face Restore.
Tile-to-Patch (TTP) Enhancements
TTP_Image_Tile_Batch
: Breaks down the image into tiles for localized Face Restore.TTP_Image_Assy
: Rebuilds a seamless image after tile-level repair, using 128px padding.
Interrogate
Joy Caption Two
: Automatically describes restored images to help validate Face Restore results.
More About This Face Restore Workflow
Based on the original technique by Xing Jiu, this workflow demonstrates how tile-based processing and identity-aware modeling can significantly improve Face Restore results on difficult image inputs.
Acknowledgements
This ComfyUI-based Face Restore workflow is adapted from the Flux TTP Tile Upscale method shared by Xing Jiu, and built using community tools like comfyui-ttp-toolset
, ky-nodes
, and easy-use
. The combination of tile patching, FluxGuidance, and LoRA integration enables professional-grade Face Restore results even on challenging inputs.