CodeFormer Online — Free AI Face Restorer
Restore old, damaged, or heavily degraded faces using CodeFormer AI from Shanghai AI Lab. Adjustable fidelity slider lets you control the balance between restoration quality and identity preservation. No Python, no download — works in your browser.
3 free uses per day · No signup required
How CodeFormer Works
The codebook-based approach to face restoration
Codebook Lookup
CodeFormer learns a discrete codebook of high-quality facial components — eyes, nose, mouth, skin textures. It maps degraded features to their closest codebook entries for reconstruction.
Fidelity Control
Unique adjustable slider lets you control quality vs. identity. Low fidelity (0.3) for maximum quality on heavily damaged photos. High fidelity (0.9) for minimal alteration of the original face.
Identity Preservation
The codebook approach inherently preserves identity better than generative methods. The restored face looks like the same person, not a generic high-quality face.
Upscale Option
Combine face restoration with upscaling (1x to 4x) in a single pass. Restore and enlarge old photos simultaneously for print-ready resolution.
CodeFormer vs GFPGAN: When to Use Each
Both models are free on UpscaleFast — here is when to choose CodeFormer
Heavily Damaged Photos
CodeFormer excels when faces have scratches, tears, water damage, or extreme blur. Its codebook approach can reconstruct faces from minimal information, where GFPGAN may produce artifacts.
Identity-Critical Restoration
When restoring photos of loved ones, identity preservation matters most. CodeFormer at fidelity 0.7-0.9 keeps the restored face recognizable while still improving quality significantly.
Very Old Photographs
Pre-1960s photos, tintype portraits, and severely faded images benefit from CodeFormer's ability to reconstruct facial structure from almost nothing. Set fidelity to 0.3-0.5 for these cases.
Fine-Tuned Control
The fidelity slider gives you granular control that GFPGAN doesn't offer. Process the same photo at different fidelity levels to find the sweet spot between quality and faithfulness.
Read our full comparison: GFPGAN vs CodeFormer: Which AI Face Restorer Is Better?
CodeFormer Online — FAQ
Yes. UpscaleFast runs CodeFormer on cloud GPUs. Upload your photo, adjust the fidelity slider, and download the restored result. No Python, no pip install, no CUDA drivers needed.
CodeFormer is an AI face restoration model developed by Shanghai AI Laboratory. It uses a learned codebook of facial components to reconstruct damaged faces. Its key innovation is the fidelity slider that lets you balance restoration quality against identity preservation.
The fidelity slider (0.0 to 1.0) controls the balance between enhancement quality and faithfulness to the original face. Low values (0.3-0.5) produce higher quality results with more AI creativity. High values (0.7-1.0) stay closer to the original face. Default 0.7 works well for most photos.
Yes, for severely damaged old photos, CodeFormer generally outperforms GFPGAN. Its codebook approach better preserves identity when the input face is heavily degraded. GFPGAN is faster and sharper for mildly blurry faces.
Yes. UpscaleFast offers 3 free CodeFormer face restorations per day with no account required. No watermarks on results.