Bring old black and white pictures to life. A neural network trained on millions of color photos predicts natural colors for skin, sky, grass and clothing, and it runs entirely inside your browser, so your family photos are never uploaded. Fine-tune the color intensity and warmth, compare before and after, and download a full-size PNG or JPG.
Last updated:
◂ ▸
OriginalColorized
Working…
grayvivid
coolerwarmer
Use your own ONNX model
Load a colorization model (.onnx) from your computer. It must take an RGB image and return RGB or two Lab color channels.
Advertisement
Advertisement
How to use the Colorize Black and White Photo
Add a photo: drop a black and white picture on the box, choose it or paste it with Ctrl+V. A sample photo is shown until you do.
Click Colorize photo. The first time, the AI model downloads (about 61 MB) and is stored in your browser; after that colorizing starts right away.
Adjust the look with the Color intensity and Warmth sliders. Changes are instant because the AI colors are kept in memory.
Compare by dragging across the photo or with the Before / after slider.
Download the result as PNG (best quality) or JPG (smaller file).
Why only the color changes, never the detail
Every pixel of a photo can be split into luminance (how bright it is) and chrominance (which color it has). A black and white photo already contains the full luminance; only the chrominance is missing. This tool works in the YCbCr color space used by JPEG:
Y (brightness) is taken from your original: Y = 0.299 R + 0.587 G + 0.114 B.
Cb and Cr (blue-difference and red-difference) come from the AI model, which works at 256 × 256 px, and are scaled smoothly up to your photo's size.
The final pixel is rebuilt with R = Y + 1.402 Cr, G = Y − 0.344 Cb − 0.714 Cr, B = Y + 1.772 Cb.
Because the human eye sees fine detail mostly in brightness and only coarse detail in color, a low-resolution color map on top of full-resolution brightness looks sharp. It is the same trick JPEG and digital TV use (chroma subsampling). The Color intensity slider simply multiplies Cb and Cr, and Warmth shifts them toward orange or blue.
What to expect from AI colorization
Subject
Typical AI result
Tip
Sky, water
Reliable blues and soft gradients
Lower Warmth slightly for a clearer sky
Grass, trees, foliage
Reliable greens
Raise intensity to 120-140% for summer scenes
Faces and skin
Usually natural, sometimes a little pale or orange
Adjust Warmth by ±10
Clothing, cars, painted walls
A plausible guess, often muted or brownish
The real color cannot be known from gray alone
Faded or low-contrast prints
Weaker color
Increase intensity; rescan with more contrast
Already colored photo
Recolored from its gray version
Useful for faded or color-cast prints
Tips for restoring old family photos
Scan prints at 300 DPI (600 DPI for small wallet photos) and save the scan as PNG or high-quality JPG.
Crop away scanner borders and white frames first with the Image Cropper; borders can tint the edges.
Reduce grain or scanner noise with the Image Denoiser, and enlarge small originals with the Image Upscaler before colorizing.
Keep the original scan. A colorized version is an interpretation; when sharing, it is good practice to say the colors were added.
Frequently asked questions
How does AI photo colorization work?
The model is a convolutional neural network (DeOldify) trained on a large set of color photos that were converted to grayscale. It learned which colors usually go with which shapes and textures, such as blue sky, green foliage and skin tones. Given your gray photo it predicts plausible colors. This tool then keeps the brightness of your original and takes only the color (chroma) from the model, so no detail is lost.
Are the colors historically accurate?
No AI can know the real color of a dress or a car in an old photo; it predicts likely colors from patterns it has seen. Skies, foliage and skin are usually convincing, while clothing and painted objects are a best guess. Treat the result as a pleasing restoration, not as a historical record.
Is my photo uploaded to a server?
No. The AI model is downloaded to your browser and runs on your own device with ONNX Runtime Web, using WebGPU when your browser supports it and WebAssembly otherwise. The photo itself never leaves your computer or phone.
Why does the first colorization take longer?
The first time, your browser downloads the model file (about 61 MB) and prepares it. It is saved in the browser cache, so later visits colorize in a few seconds. Processing time also depends on your device; a recent laptop usually takes 2-10 seconds per photo.
What is Quick tint?
Quick tint is a simple, non-AI method that maps each gray level to a color from a hand-tint palette (dark shadows, warm skin and earth midtones, light highlights) and adds blue to bright, smooth areas near the top of the frame where sky usually is. It is instant and works offline, but it cannot recognize objects, so use the AI mode for realistic results.
What photos give the best results?
Sharp scans with good contrast work best. Scan prints at 300-600 DPI, crop away borders, and fix heavy scratches first if you can. Very faded or very dark photos colorize less convincingly; raising the color intensity a little can help.
Can I colorize a photo that already has color?
Yes. The tool converts it to black and white first and then adds new colors, which can revive a badly faded or color-cast print. The original colors are not reused.
Are all the tools on FreeToolCabin tested for quality and accuracy?
Yes! Every tool on our site undergoes automated Q/A testing in real browser environments to ensure high precision, correct output, and zero errors. If you happen to spot a bug or have a suggestion, please contact us at [email protected] and our team will review it right away.