🧠 ComfyUI Face Model Pipeline — Powered by ReActor | Build Perfect Face Models in 3 Clicks

Stop wasting hours on manual face swaps. This workflow builds production-ready face models from a photo batch — automatically.

✨ What This Workflow Does

Feed it 20+ photos of any person. It automatically:

• Detects & crops every face with sub-pixel precision

• Upscales each crop to 512×512 using Bicubic interpolation

• Extracts deep face embeddings via ReActor

• Saves a ready-to-use .safetensors face model directly to your InsightFace folder

One click. One model. No manual work.

🚀 Why This Workflow Crushes Everything Else

Feature This Workflow Manual Method Setup time~5 minutes Hours Face extraction Auto-batch One by one Output format.safetensors (ReActor-ready)Varies Face alignment✅ Automatic❌ Manual Batch processing✅ Full folder❌ Image by image Node layout Clean, aligned grid Spaghetti chaos

📦 What's Included

• FaceModel_Pro.json — Drop-and-go ComfyUI workflow

• Full EN Documentation (10-section professional guide)

• README with Quick Start & Tips

• CHANGELOG with full version history

• LICENSE — Clear personal & commercial use terms

⚙️ Requirements

Extensions (install via ComfyUI Manager):

• ComfyUI-ReActor

• WAS Node Suite (was-ns)

System:

• Python 3.10+

• ComfyUI (latest)

• CUDA 11.8+ recommended

• 8 GB RAM (16 GB recommended)

• ~500 MB disk for models

🎯 Quick Start — 3 Steps

Step 1 — Install the two extensions above via ComfyUI Manager

Step 2 — Drag & drop FaceModel_Pro.json onto your ComfyUI canvas

Step 3 — Set your image folder path → set your model name → hit Queue Prompt

✅ Your face model saves to: ComfyUI/models/insightface/

💡 Pro Tips for Best Results

• Use 20+ images minimum — more photos = stronger, more consistent model

• Good lighting is everything — avoid heavy shadows or backlit shots

• One face per frame — multi-face images confuse the detector

• Increase crop_padding_factor to 0.35–0.5 if you want neck/shoulders included in the crop

• Consistent expressions across your dataset improve embedding quality

🔧 Workflow Defaults

crop_padding_factor : 0.25
upscale_method      : bicubic
output_resolution   : 512 × 512
save_mode           : ON
select_face_index   : 0
batch_mode          : incremental_image

All values are fully adjustable inside ComfyUI.

🗺️ Roadmap — Coming in v1.1.0

• Live Preview Image node before saving

• Multi-face export per batch run

• LANCZOS upscale option

• Face quality filter — auto-skip blurry or low-res detections

• Arabic documentation

❓ FAQ

Q: Does this work with anime faces? A: Detection uses lbpcascade_animeface.xml — yes, anime faces are supported!

Q: Can I use the generated model commercially? A: Check the included LICENSE.txt for full usage rights.

Q: What if no face is detected? A: Make sure images are well-lit and front-facing. Increase crop_padding_factor slightly and retry.

Q: How many photos do I actually need? A: 20 is the minimum. 40–60 clean, varied photos will give you a noticeably stronger model.