Old photo restoration
family albums, historical archives
// official site: github.com ↗
AI / LLM · PRO TIER
GFPGANpro
GFPGAN is Tencent ARC's practical face restoration algorithm — restores old, blurry, or low-resolution face photos using generative adversarial network priors. Trained on millions of high-quality face images, GFPGAN handles faces from baby to elderly with consistent quality.
// What it is
A closer look.
GFPGAN is Tencent ARC's practical face restoration algorithm — restores old, blurry, or low-resolution face photos using generative adversarial network priors. Trained on millions of high-quality face images, GFPGAN handles faces from baby to elderly with consistent quality.
The industry standard for AI face recovery — the "before/after" you've seen in viral Reddit posts comes from GFPGAN.
// Use cases
What it's for.
Concrete scenarios where teams pick GFPGAN over the SaaS alternative.
Blurry photo enhancement
social media, scanned docs
Low-resolution profile pictures
to high-res
Document face cleanup
passport photos, ID cards
AI-generated image fix
repair SD/Flux face artifacts
// Who it's for
Built for these teams.
If your team profile matches one of these, GFPGAN is a strong fit out of the box.
Photographers
restoring old client photos
Family archivists
digitizing/enhancing old albums
Document workflow teams
cleaning passport/ID scans
AI art creators
post-processing SD face artifacts
Marketing teams
enhancing low-res legacy images
// Differentiators
Why teams pick GFPGAN.
When evaluating self-hosted options for this category, here are the dimensions on which GFPGAN consistently lands above the alternatives.
- ✓Apache 2.0 — fully open
- ✓Best face-specific restoration — in the open-source space
- ✓Robust to severe degradation — (blur, noise, JPEG artifacts)
- ✓Multiple version variants — for different aesthetics
- ✓Integrates with Real-ESRGAN — for full-image restoration
- ✓Used in A1111 / ComfyUI — natively
- ✓Tencent ARC backing
// Integrations
Connects to.
The stack you'll plug GFPGAN into — services, protocols, and adjacent apps in the BluixApps catalog.
◇
Gradio web UI
(BluixApps custom)
◈
CLI mode
for batch processing
◆
A1111 Extras
built-in
▣
ComfyUI nodes
restoration node pack
▦
Pair with
Real-ESRGAN (Real-ESRGAN for background + GFPGAN for faces = best total restoration)
▩
Pair with
SD/Flux generation (ADetailer-like post-process)
// Adoption & deployment
Notable users & community
- 37k+ GitHub stars
- Tencent ARC research backing
- Standard in commercial photo-AI products
- Featured in viral restoration before/after posts
- Multiple language tutorials + courses available
What we ship
- Cloned
TencentARC/GFPGANrepo - pytorch CUDA 12.4 base
- Custom Gradio UI with version + scale selectors
- Persistent volumes: repo, weights (~700 MB), input, output
- Port 7876 mapped
- Install report at
/root/bluixapps/gfpgan.txt - Version selection guide
- CLI batch examples
- Use case examples (photo archive, document scanning, AI art fix)
- Pairing suggestions (Real-ESRGAN combo for best results)
- GPU pre-flight check via
bluixapps_ensure_nvidia_runtime - Backup hook covers weights + outputs
// Tips & operations
Run it properly.
Operational guidance from running this in production — what to lock down, what surprises people.
// PERFORMANCE
Versions
// SECURITY
Upscale factor
1-4× (combined with face restore)
// OPERATIONS
VRAM
4 GB minimum
// RELIABILITY
Speed
1-3 sec per photo on RTX 4090
// DEPLOYMENT
Best inputs
degraded photos (blur, low-res, JPEG artifacts) — that's what GFPGAN learned to fix
// SCALING
Combine with Real-ESRGAN
--bg_upsampler realesrgan for full image enhancement
// MAINTENANCE
Batch CLI
process whole folders for archive work
6144
// min ram (MB)
6
// min disk (GB)
7876
// access port
http
// protocol
pro
// bluixapps tier
// Alternatives in AI / LLM
Compare with
Project resources
Official sitegithub.com ↗