PixelOkay
Photos stay local
Tools
Tools›Images›AI Photo Colorizer
AI PHOTO STUDIORESTORE THE FEELING

AI Photo Colorizer

Recolor photographs manually with editable local color layers. Optional on-device DDColor can help after you choose to enable it.

✦ No AI required✦ Color layers & region masks✦ High-quality export
YOUR WORKSPACE

Colorize your photo ✳

Import, colorize, refine, and save. Every photo stays in your browser.

01

Source photograph

Start with a scanned or digital photo

⇧Drop a photo hereor click to browse filesPNG, JPEG, WebP · up to 25 MB
No photo selected
02

Choose your approach

Manual editing works immediately. AI is optional.

First run downloads a free ~135 MB model and ONNX runtime. This requires an internet connection; inference runs locally. The loaded model is reused for further AI applies, but reset or cancel releases it. Some browsers and low-memory devices may not support it. AI colors are plausible guesses, not historically verified.

03

Fine-tune the color

Adjust tone and chroma without any model

100%
100%
0
0
0
0
Manual color correctionsApply natural-looking color while preserving detail.
5%
70%
20%
Color layersUp to 8 · erase affects this layer only
100%

Paint or draw regions on the photo. Choose a color layer, then adjust its opacity. Color is an artistic interpretation, not historical evidence.

LIVE PREVIEW

See the transformation

▧
Your photo will appear hereChoose an image or try the sample to begin.
100%
Ready to import
EXPORT

Save the finished photo

Local export

Project JSON includes edits and a photo reference, not the photo itself. Keep the original to reopen.

Export preserves original luminance at the chosen processing resolution (up to 2560 px long edge / 6 MP). Original files are never overwritten.

✦
Tip: Correct colors with the brush

Automatic colorization may guess clothing, walls, or landscapes incorrectly. Enable the brush to gently paint more accurate hues without flattening the photographic detail.

HOW IT WORKS

History deserves more than a filter.

✳

Local AI color prediction

A free DDColor-tiny model predicts color channels while your original image keeps its fine details. The model downloads only when you choose Download & Load Model.

◌

Hands-on artistic control

Adjust color intensity, warmth, tint and contrast, then paint corrective colors on a layer. Undo and redo your corrections as you refine.

◇

Private by architecture

Photos and generated pixels are processed locally. The browser fetches open-source runtime and model assets, but the tool does not send your image to a server.

Frequently asked questions

Does this recover the real colors from the past?

No. AI colorization predicts plausible colors; it cannot establish the historical truth of clothing, hair, or scenery. For archival work, record edits as interpretations and retain the original.

Do I have to pay for the AI model?

No. The DDColor-tiny ONNX model is available under Apache-2.0. The first run needs an internet connection to download approximately 135 MB of model data; later loading may benefit from the browser cache.

Why does AI colorization sometimes fail?

Large neural models require substantial memory and WebAssembly support. Restricted networks, insufficient RAM, unsupported browsers, or blocked downloads can cause failures. Manual coloring and corrections remain available without the model.

Are my photographs uploaded?

No photo upload endpoint is used. Your browser downloads external open-source code and model files when AI colorization is requested, then processes your photo on your device. Review network policies before using any browser tool with sensitive material.

Can I use this to recolor already colored images?

Yes. Automatic mode intentionally derives a grayscale luminance channel and predicts new chroma. Manual correction can also recolor regions. Your source photo is retained for comparison and reset.

LOCAL AI SETUP

Set up AI on this device

Load DDColor-tiny on your device

Photo Colorizer runs DDColor-tiny ONNX inside a local Web Worker. Your photo pixels are not uploaded. Only after you choose Download & Load will the browser fetch the approximately 135 MB model and ONNX runtime.

  1. 1. Check device
  2. 2. Download / cache
  3. 3. Load ONNX model
  4. 4. Colorize photo
DDColor-tiny FP16 · 135.4 MB pinned model · ONNX Runtime Web 1.21.0 · WASM/CPU. No WebGPU requirement. Colors are plausible estimates, not historical facts.

Device & model resources

Measurements may be approximate
Device RAM / model memoryChecking device information…

Live system RAM, model RAM and GPU memory are not reliably exposed in browsers.

Browser website storageChecking storage…

Site-wide estimate; includes other PixelOkay tools and is not RAM.

ONE ACTION · ALL RECOGNIZED AI MODEL DOWNLOADS

Free AI model storage across PixelOkay

Scans same-origin model Cache Storage (including recognizable WebLLM, Transformers.js and ONNX model files from other PixelOkay tools). No browser databases, projects, imported photos, or site data are cleared.

Run a scan to see how many model caches can be cleared together.

Browser-native models and unknown IndexedDB/OPFS or private runtime caches may remain. Unknown application/user storage will not be deleted. Other running tools might redownload caches; close their tabs first.

Having trouble downloading or loading AI?

Old or incomplete AI model downloads may be using your browser's storage. First refresh the storage meter and retry the current compatible model. If setup still fails, inspect and remove recognized AI model caches with confirmation, then download again.

Only model caches are targeted. Other PixelOkay tools may need to redownload their models. Never clear all website data without exporting important chats and projects first.

Select “Check device” to see whether the current model can run.
ONNX/WASM is required. WebGPU is not required. Runtime downloads begin only after user approval.

First colorization may take longer while ONNX compiles its CPU kernels. Processing runs in a Worker; actual first-use responsiveness varies by device.
CROSS-TOOL MODEL CACHE CLEANUP

Delete all recognized AI model downloads?

Deleting these caches will cause other PixelOkay tools to download their models again. Loaded AI sessions in other tabs may be disrupted. Unidentified model caches and browser-managed models may remain.

Data-loss caution: This action does not intentionally delete saved chats, projects, drafts, localStorage or IndexedDB. Do not clear website data unless you have backed up work; clearing all website data may permanently erase saved work.

Using AI Photo Colorizer

Open a supported photograph or the sample, apply optional AI colorization, and refine colors with the manual brush and tone controls. Compare the result and export at the working resolution. Manual editing is available independently of the model.

AI downloads a large DDColor model plus ONNX Runtime. Model colors are estimates, not recovered historical colors; the configured byte-count check is not a cryptographic hash. Working copies are limited to 2560 pixels on the long side and 6 MP. Device memory can impose lower limits. JPEG uses a white matte; clipboard copying needs browser permission. Exported pixels do not retain original metadata.

References: DDColor model · ONNX Runtime

Workflow and output choices for AI Photo Colorizer

Open a supported photo or the supplied example, request optional DDColor colorization, then refine selected areas with the manual brush and tone controls.

Exports use the bounded working resolution and do not retain original image metadata. The model proposes color estimates; manual edits are independent of the model.

Example workflow

  1. Open a supported grayscale photograph and review its working-resolution preview.
  2. Request AI colorization if desired, then compare the result with the original and correct regions with the manual brush.
  3. Export the reviewed image and retain the source photograph separately.

Authored example: Use an authored example of a grayscale garden photo and mark one region where an AI color estimate should be reviewed manually. Do not label a chosen hue as historically verified.

Limits and model handling: AI requests a large DDColor model and ONNX Runtime from external providers. Colors are estimates, not recovered historical facts; the configured byte-count check is not a cryptographic hash. Working images are limited to 2560 pixels on the long side and 6 MP.

AI Photo Colorizer questions

Does colorization recover the photograph’s original colors?

No. Model output is an estimate and cannot establish historical colors. Review and correct it manually where useful.

Does manual color editing require the AI model?

No. The brush and tone controls can be used independently; the model is requested only for AI colorization.

Related tools