Free · In-Browser DDColor AI · 100% Local & Private
Restore scratched or faded black-and-white photos and colorize them directly in your browser using the Alibaba DAMO DDColor neural network. 100% local privacy, full-resolution CIE Lab luminance locking, 7 film presets, and live 1:1 split comparison.
Run a real 110 MB deep neural network (DDColor-Tiny) directly inside your browser via WebAssembly SIMD & WebGPU. Suppress scratches, sharpen facial features, and reconstruct lifelike colors while keeping 100% of your original pixel resolution and privacy.
Traditional cloud tools compress your photos and risk privacy leaks. UTools brings Alibaba DAMO Academy’s dual-decoder DDColor neural architecture directly into your browser with zero-loss CIE Lab luminance locking.
Combines edge-preserving unsharp masking with DDColor cross-attention queries to restore natural warm skin tones, subtle cheek warmth, and crisp eye catchlights.
Automatically distinguishes high-luminance sky gradients from mid-tone foliage and architectural stone, eliminating color bleeding across object boundaries.
Instead of replacing your image with a low-res AI output, we lock 100% of your original high-resolution lightness (L*) and fuse bilinearly upsampled neural chromaticity (a*, b*).
| Dimension | ✨ UTools In-Browser AI (DDColor + YCbCr) | Traditional Cloud Upload Tools |
|---|---|---|
| Privacy & Data Security | ✅ 100% local in browser; family photos never leave your device | ❌ Must upload private family photos to remote third-party servers |
| Output Image Resolution | ✅ CIE L* channel locked at 100% original resolution (up to 2048px) | ❌ Often downsampled to 512px or heavily JPEG-compressed unless paid |
| AI Model Architecture | ✅ Alibaba DAMO DDColor Dual-Decoder Neural Net + 7 Film Presets | ⚠️ Black-box fixed output with no interactive warmth/saturation tuning |
| Cost & Usage Limits | ✅ 100% Free, unlimited photos, no watermark, works offline once cached | ❌ Strict paywalls, queue waiting times, and watermarks on exports |
Upload 1 to 5 old photos (PNG, JPEG, WebP up to 15 MiB) or click one of the 3 built-in vintage sample photos (1940s Portrait, 1950s Landscape, 1960s Street). Switch between the Instant YCbCr Semantic Engine (0-ms response) and the In-Browser DDColor AI Neural Network (auto-cached in browser CacheStorage). Choose from 7 quick presets (Portrait Colorize, Vivid Scenery, Vintage Kodak, Historic Documentary, Denoise & Crisp, Subtle Touchup, Antique Sepia), fine-tune scratch reduction, sharpness, saturation, and warmth, then drag the 1:1 stationary split slider and download the full-resolution JPEG.
AI colorization and scratch suppression reconstruct plausible chromaticity from grayscale luminance cues rather than historical ground truth. All pixel processing and ONNX neural inference execute 100% locally inside your browser; photos are never uploaded to any external server. The 110 MB DDColor ONNX model is downloaded once and persisted in browser CacheStorage for offline reuse.
Click the built-in “1940s Vintage Portrait” sample or upload a scanned family portrait, switch to “In-Browser AI Deep Neural Net (DDColor)”, select the “Portrait Colorize” or “Vintage Kodak” preset, and drag the stationary split slider to inspect lifelike skin, lip, and background restoration before exporting.
Stage 1 applies 3×3 edge-preserving median filtering to suppress dust and scratches, 1%–99% adaptive histogram stretching to remove yellowing haze, and Laplacian unsharp masking for facial clarity. Stage 2 decouples the image in CIE D65 L*a*b* color space, locking 100% of the full-resolution lightness channel (L*) so fine textures never blur. Stage 3 predicts chromaticity (a*, b*) via either the instant multi-zone YCbCr semantic engine or Alibaba DAMO Academy’s DDColor-Tiny dual-decoder neural network (ConvNeXt visual backbone + multi-scale Transformer cross-attention color queries) executed locally via ONNX Runtime Web WASM SIMD, then bilinearly upsamples chroma and recombines with the original L* channel.
Restore scratched or faded black-and-white photos and run the Alibaba DAMO DDColor neural network directly in your browser. Preserve 100% original resolution via CIE Lab luminance locking with 7 film presets and 1:1 split comparison.
Upload 1 to 5 old photos (PNG, JPEG, WebP up to 15 MiB) or click one of the 3 built-in vintage sample photos (1940s Portrait, 1950s Landscape, 1960s Street). Switch between the Instant YCbCr Semantic Engine (0-ms response) and the In-Browser DDColor AI Neural Network (auto-cached in browser CacheStorage). Choose from 7 quick presets (Portrait Colorize, Vivid Scenery, Vintage Kodak, Historic Documentary, Denoise & Crisp, Subtle Touchup, Antique Sepia), fine-tune scratch reduction, sharpness, saturation, and warmth, then drag the 1:1 stationary split slider and download the full-resolution JPEG.
The Instant YCbCr Semantic Engine requires zero model download and responds in 0 ms for quick scratch removal and film preset previews. The In-Browser DDColor AI Engine executes Alibaba DAMO Academy’s 110 MB dual-decoder neural network (ConvNeXt + Transformer) locally inside your browser to semantically recognize faces, sky, foliage, and clothing for studio-grade colorization.
No. 100% of pixel restoration and DDColor ONNX neural network inference run locally inside your browser sandbox. Zero bytes of your family photos are ever uploaded to any external server.
Traditional AI tools output raw low-resolution neural reconstructions that blur facial details. UTools decouples lightness and chroma in CIE D65 L*a*b* space, locking 100% of your original full-resolution luminance channel (L*) and fusing bilinearly upsampled AI chromaticity (a*, b*), guaranteeing zero sharpness loss.
No. After the first silent background preload or initial run, the 110 MB ONNX model is permanently persisted in your browser’s local CacheStorage. All subsequent visits load the neural network from local disk in milliseconds with zero network traffic.
AI colorization and scratch suppression reconstruct plausible chromaticity from grayscale luminance cues rather than historical ground truth. All pixel processing and ONNX neural inference execute 100% locally inside your browser; photos are never uploaded to any external server. The 110 MB DDColor ONNX model is downloaded once and persisted in browser CacheStorage for offline reuse.