AI Old Photo Restorer
Restore old, faded, or damaged photos free in your browser using the Swin2SR Image Restoration Transformer — a neural network from Microsoft Research trained specifically on degraded images. It repairs noise, JPEG artifacts, and lost detail entirely on your device. 100% private, no signup, no watermark. Works offline after first model load.
POST-PROCESSING STRENGTH
Drop your old photo here or click to upload
Supports JPG, PNG, WebP — max 10MB
Restored Photo
How Swin2SR AI Photo Restoration Works
This AI Old Photo Restorer uses Swin2SR — a Swin Transformer V2-based image restoration model from Microsoft Research — running via Transformers.js in your browser. The model was trained specifically on compressed and degraded images, making it ideal for old scanned photographs. Unlike simple Canvas sharpening, Swin2SR understands image structure and reconstructs fine detail that was genuinely lost — faces, textures, edges — rather than just boosting contrast. The model is downloaded once (~50MB) and cached in your browser for offline use. Last updated: March 2026.
The restoration pipeline: your photo is resized to an optimal processing size, fed through the Swin2SR neural network, and output at 4× the resolution with reconstructed detail. A post-processing pass then applies contrast and sharpening at the strength level you select (Light / Standard / Intensive). If the AI model fails (very old browser or slow connection), the tool automatically falls back to Canvas-based smart processing.
Restore Old Family Photos in Seconds
The most common use is restoring old family photographs — grandparent portraits from the 1940s–1970s, faded colour prints from the 1980s–1990s, or black and white scans from the early 20th century. Professional restoration services charge $30–$200 per photo. This tool is completely free and processes in under 30 seconds. For black and white photos, use this restorer first to improve clarity, then use the AI Photo Colorizer to add realistic colour. For heavily blurry portrait photos, also try the AI Face Restorer for portrait-optimised enhancement.
What the AI Fixes
- JPEG and scan compression artifacts: Blocky, pixelated regions from lossy scanning — Swin2SR reconstructs the original smooth detail.
- Noise and grain: Film grain and scanner noise are suppressed while preserving real edge detail.
- Lost fine detail: Hair, fabric texture, facial features — the neural network fills in high-frequency content lost to age and scanning.
- Fading and colour cast: Post-processing contrast correction addresses tonal compression and yellowing from age.
Before and After: Real Restoration Results
The difference is most visible on compressed or low-resolution scanned photos. A scanned 1960s portrait at 640×480 pixels is processed through Swin2SR and output at 2560×1920 with reconstructed detail. Faces, clothing, and background elements that appeared as blurry smudges in the original become legible. The neural network does not invent content — it reconstructs what the compression and degradation removed, based on learned patterns from thousands of degraded-to-clean image pairs.