Image Colorizer

Add realistic color to black & white, grayscale, and vintage photos with AI. DeOldify deep learning predicts historically plausible colors for faces, landscapes, and everyday scenes. Run it free to preview the result; download the clean full-resolution image with credits.

Artistic: more vibrant colors. Stable: more natural, better for portraits and landscapes.

35
Faster Higher quality

Drag & drop a file here, or

JPG PNG WebP Max 20MB • Processed on server • Deleted immediately

AI colorization engine

Uses DeOldify, a GAN-based deep learning model trained on millions of images to predict realistic colors for grayscale input. Two modes: Artistic for vivid results, Stable for natural skin tones and landscapes.

Two Colorization Modes

Artistic mode produces the most vibrant, eye-catching colors — great for sharing. Stable mode is more conservative, preserving natural tones and avoiding over-saturation in faces and nature scenes.

Private & Secure

Your photo is sent to our AI server, colorized there, and deleted immediately after processing. It is never shared or used for anything else. The clean full-resolution result is held only briefly so you can download it, then removed.

How it works

How AI photo colorization works

This tool uses DeOldify, a deep learning model based on a Self-Attention GAN with NoGAN training. It analyses a grayscale image and predicts plausible RGB colors for every pixel — skin tones, sky gradients, foliage, fabrics, metal, wood. The model was trained on millions of natural images from ImageNet, so it learns what colors real-world objects tend to be. It runs on our AI server in 5-15 seconds, and the original photo is deleted immediately. Historical accuracy is a known challenge — the AI guesses likely colors, not what was actually there. To inspect a photo's metadata before colorizing, use the EXIF Checker.

Artistic vs Stable — which model to choose

DeOldify ships two image models, each with different strengths. Artistic uses a ResNet34 U-Net with extra decoder depth and five NoGAN training cycles — it produces the most vivid, eye-catching colors but can occasionally leave parts of an image gray or oversaturate. Stable uses a deeper ResNet101 backbone with more conservative training — it is more consistent with skin tones, nature scenes, and portraits, though colors may be less dramatic. Start with Artistic for social sharing and editorial use; switch to Stable if faces look wrong or landscapes seem over-colored. If the colorized result needs resizing afterward, try the Image Resizer or Social Resizer.

Colorize vs restore — what's the difference

Colorization adds color to grayscale images — it does not fix blur, damage, or low resolution. If your old photo is both black-and-white and blurry, colorize first, then run the result through the Photo Unblur to recover edge detail. Scanly runs no face model, so faces are never reconstructed. If it only needs to be made bigger, the Image Upscaler (Real-ESRGAN) is the better fit. For background removal from the colorized portrait, use the Background Remover.

Common use cases for image colorization

Bringing old family photos to life for framing, gifting, or sharing online. Colorizing historical photographs for articles, documentaries, and school projects. Adding color to scanned negatives or archive prints. Enhancing black-and-white portraits for genealogy records. Previewing how a grayscale illustration would look in color. After colorizing, optimize file size with the Image Compressor, convert formats with the Format Converter, or check metadata with the EXIF Remover. Explore all 92 image tools.

Frequently Asked Questions

Is this image colorizer free?

Yes, completely free with no watermarks, no account required, and no daily limits. Processing runs on our AI server using DeOldify — your image is deleted immediately after the colorized result is returned.

What is DeOldify?

DeOldify is a deep learning model that uses a Self-Attention GAN with a novel NoGAN training technique to colorize black and white images. It was trained on millions of natural images and can predict realistic colors for people, landscapes, buildings, and everyday objects.

What is the difference between Artistic and Stable?

Artistic mode uses a ResNet34 backbone and produces the most vibrant, eye-catching colors — best for dramatic results and social sharing. Stable mode uses a deeper ResNet101 backbone and is more conservative — it handles skin tones, portraits, and nature scenes more consistently, with fewer gray patches. Try Artistic first; switch to Stable if faces look off.

What does the Render Quality slider do?

The render quality (render factor) controls the resolution at which the color portion of the image is processed. Higher values produce more detailed and vibrant color but take slightly longer. Lower values are faster but may produce less saturated results. The default is 35 for Artistic and 25 for Stable — these work well for most images.

Are the colors historically accurate?

Not necessarily. The AI predicts plausible colors based on patterns learned from modern photographs — it does not know the actual original colors. Clothing, backgrounds, and fine details are best guesses. The results are realistic but should not be treated as ground truth for historical research.

Can I colorize a photo and then sharpen it?

Yes. If your old photo is both black-and-white and blurry, colorize it first with this tool, then upload the colorized result to the Photo Unblur to recover edge detail, or to the Image Upscaler to enlarge it. Scanly runs no face model, so faces are never reconstructed or altered.

Are my images stored on the server?

No. Your image is sent to our AI server, colorized there, and the result is returned to your browser. The original image is deleted immediately after processing — nothing is stored, logged, or shared.

Scanly is an independent product. DeOldify is a third-party model we call through Replicate — an integration, not our brand. Scanly is not affiliated with, endorsed by or sponsored by Replicate or the model authors, and all model names and trademarks remain the property of their respective owners. AI Disclosure.