Numa 0.26.2

Numa for Linux, as a Flatpak. Install with flatpak install --user Numa-0.26.2-x86_64.flatpak; the GNOME 50 runtime downloads the first time.

Or the AppImage: chmod +x Numa-0.26.2-x86_64.AppImage and run it.

0.26.2

• The camera profile picker in the Colour tab lists your camera's profiles again.

0.26.1

• Sliders stay smooth in large libraries: dragging one no longer makes the whole window lay itself out again on every step.

0.26.0

• Lens corrections now run on Numa's own code, reading the LensFun database, and find more lenses and cameras by the names in their files. The Colour tab has a camera profile picker again, with Numa's own profiles for six cameras.
• Twelve built-in looks, with a Strength slider once one is chosen. LUTs (.cube and .3dl) with an Amount. A waveform and RGB parade beside the histogram.
• A Canon multiple exposure no longer comes out magenta.

0.25.1

• A JPEG, HEIF or PNG opens looking exactly as it is: Numa no longer lays its raw contrast and sharpening over a finished picture, and a Display P3 or Adobe RGB photograph keeps its colours.

0.25.0

• A library on a drive that is not connected stays in the list and opens from what Numa remembers: the thumbnails it already made, each marked as away. Stars and flags given meanwhile go into the library when the drive is back.
• The grade has a colour wheel for each range, as on the iPad.

0.24.2

• On the iPhone and iPad, selecting by a tap asks for a quarter of the memory it did and answers faster, with the same selections.

0.24.1

• The mask models give their memory back after each use instead of keeping their highest mark, so an editor with masks holds about a gigabyte less.

0.24.0

• A night of testing: a folder renamed inside a library keeps its ratings and edits, undo and history steps are saved at once, and TIFF and WebP files open. Subject and Person no longer hold up the window, the mask bar fits a narrow window, the light scheme is readable, a quarter turn keeps the crop and its ratio, and AVIF exports carry the creator, copyright, rating and keywords.

0.23.0

• Subject is what the photograph is about, found by BiRefNet, and is a mask of its own rather than the person another model found. Person stays as it was. Heads, hats and feet are no longer cut through, and Refine edge keeps thin things.
• Clicking on a thing follows the photograph's edge instead of blocks, with far fewer stray spots and holes. The groups are cleaner: Mountains is new, and sky between branches is Sky.

0.22.0

• Subtract: a person, an animal, the sky or anything else the model finds can be cut back out of a mask — the gradient over the sky with the person on the horizon left as they were. A person or an animal is cut along the matting model's edge.
• The export's watermark is set by the same code on every platform. Where the processor runs AI sharpen through Core ML, it now runs at one tile size, which lets the whole model through.

0.21.0

• A mask can do everything the photograph can that belongs to a region: the colour mixer and black and white, its own point colours and colour grade, red, green and blue curves, a Color that washes the region in one hue, grain, and all of Detail.
• AI denoise and AI sharpen are buttons rather than switches. A point colour just picked is scrolled to and lit up, and its hue slider is in colour like the mixer's. Masks are kept on disk, so a photograph with a subject mask opens without running a model.