Severely degraded face restoration
beyond GFPGAN capability
// official site: github.com ↗
AI / LLM · PRO TIER
CodeFormerpro
CodeFormer is robust blind face restoration via codebook lookup (NeurIPS 2022) — more aggressive recovery than GFPGAN, designed for severely degraded faces. Handles AI-generated artifacts, very old photos, and other tough cases that GFPGAN struggles with.
// What it is
A closer look.
CodeFormer is robust blind face restoration via codebook lookup (NeurIPS 2022) — more aggressive recovery than GFPGAN, designed for severely degraded faces. Handles AI-generated artifacts, very old photos, and other tough cases that GFPGAN struggles with.
The alternative to GFPGAN when GFPGAN isn't enough — pick by photo, try both.
// Use cases
What it's for.
Concrete scenarios where teams pick CodeFormer over the SaaS alternative.
Color photo restoration
from B&W
AI-generated image face fix
SD/Flux face artifact compensation
Document face cleanup
passport, ID photos
Historical photo restoration
// Who it's for
Built for these teams.
If your team profile matches one of these, CodeFormer is a strong fit out of the box.
Photo restoration professionals
(commercial / archival)
Document AI teams
with degraded inputs
AI art studios
fixing generation artifacts
Historical research teams
restoring archival photos
Family genealogy services
restoring old photos
// Differentiators
Why teams pick CodeFormer.
When evaluating self-hosted options for this category, here are the dimensions on which CodeFormer consistently lands above the alternatives.
- ✓S-Lab License — academic + commercial OK with attribution
- ✓More aggressive — than GFPGAN — recovers from worse degradation
- ✓Fidelity slider — control quality vs identity preservation tradeoff
- ✓Robust to AI artifacts — fixes SD face issues better than GFPGAN
- ✓NeurIPS 2022 — publication — peer-reviewed quality
- ✓Active maintenance — + community
// Integrations
Connects to.
The stack you'll plug CodeFormer into — services, protocols, and adjacent apps in the BluixApps catalog.
◇
Gradio web UI
(BluixApps custom)
◈
CLI mode
for batch
◆
A1111 Extras
alternative to GFPGAN
▣
ComfyUI nodes
community wrappers
▦
Pair with
Real-ESRGAN (background) + CodeFormer (faces)
▩
Try both GFPGAN + CodeFormer
results vary per photo
// Adoption & deployment
Notable users & community
- 17k+ GitHub stars
- Nanyang Technological University (S-Lab) research backing
- Featured in photo restoration tooling
- Active community + tutorials
- Multiple commercial restoration workflows use it
What we ship
- Cloned
sczhou/CodeFormerrepo - pytorch CUDA 12.4 base + opencv
- Custom Gradio UI with fidelity slider + upsample toggles
- Persistent volumes: repo, weights (~600 MB), input, output
- Port 7877 mapped
- Install report at
/root/bluixapps/codeformer.txt - Fidelity guidance + use cases
- CodeFormer vs GFPGAN comparison
- CLI batch examples
- Pairing suggestions (Real-ESRGAN combo for full restoration)
- 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
Fidelity slider
// SECURITY
Background + face upsample
enable for full enhancement
// OPERATIONS
VRAM
4 GB minimum
// RELIABILITY
Speed
1-3 sec per photo
// DEPLOYMENT
Best inputs
severely degraded (blur + noise + low-res combined)
// SCALING
CodeFormer vs GFPGAN
// MAINTENANCE
Production batch
CLI mode for archive workflows
6144
// min ram (MB)
6
// min disk (GB)
7877
// access port
http
// protocol
pro
// bluixapps tier
// Alternatives in AI / LLM
Compare with
Project resources
Official sitegithub.com ↗