youcutai
DAMO Academy (open source) via fal.aiedit

DDColor: Black-and-White Photo Colorization

DDColor is an open-source image colorization model from Alibaba DAMO Academy that uses dual decoders, one for pixel-level color and one for semantic color queries, to produce vivid yet plausible colors. It performs especially well on historical photographs, assigning natural skin tones, skies and foliage. youcut.ai runs DDColor after CodeFormer restoration in the Colorize and Restore + Colorize modes.

Strengths

  • • Vivid, saturated but plausible colors
  • • Strong on faces, skies and vegetation
  • • Fast (1–2 s)
  • • Open source

Limitations

  • • Cannot know true colors of clothing or objects
  • • Occasional color bleeding across object edges
  • • Sepia input should be desaturated first (handled automatically)

Best for

  • • Family archives and historical photos
  • • Museum and education content
  • • Black-and-white film scans
Specs
TaskImage colorization
InputGrayscale or sepia, up to 2048 px
Latency~1–2 s
Provider keyfal-ai/ddcolor

Try DDColor in

Frequently asked questions

Are the colors historically accurate?+

Skin, sky and nature are highly plausible. Specific colors of uniforms, cars or dresses are informed guesses, since the original color is not recorded in the photo.

Should I restore before colorizing?+

Yes. Scratches and noise confuse colorization. The Restore + Colorize mode runs CodeFormer first and then DDColor.

Can I adjust the colors afterwards?+

Send the result to the AI Photo Editor and describe the change, e.g. "make the dress blue".

Does it work on partially colored photos?+

Faded color photos are better served by the Restore mode, which recovers contrast and saturation rather than re-inventing color.

What does colorization cost?+

2 credits for Colorize, 3 credits for Restore + Colorize.