Editing images with machine learning

October 16, 2018 — September 2, 2024

computers are awful
generative art
making things
photon choreography

Assumed audience:

No-one, now. This is comically outdated.

Why are we doing anything manually any more? There is AI to do all the menial repetitive menial stuff for images, and manual repetitive image processing is the unambiguous great triumph of modern AI.

Much of this is outdated, since it predates the stable-diffusion explosion. For bleeding-edge image-diffusion stuff see the appropriate page.

Figure 1

There is a brief efflorescence of quirky small ML companies producing these apps before they get bought up by Adobe.

1 Background removal

Background removal is something I need in practice often. Clipping Magic is fun and removes backgrounds using AI plus also a nicely-designed user interface.

Figure 2

Competitor removebg does something similar, and slightly cheaper, although there is less user configurability.

croppola does content-aware cropping.

2 cleanup.picture

3 Letsenhance

Famously viral in recent times, Let’s Enhance uses AI super-resolution and colour enhancement to make images bigger and shinier. Might also salvage something usable from facebook’s savage compression and downsampling.

4 Vance

Vance AI Image Sharpener (Pricing Plan).

5 Hotpot

Art Personalization ( API Test Tool).

6 Miscellaneous GIMP addons

Kritik Soman’s GIMP-ML plugins bring some cute ML tricks to open source image editor GIMP. I love the idea, but it is quite unwieldy.

7 References

Soman. 2020. GIMP-ML: Python Plugins for Using Computer Vision Models in GIMP.” arXiv:2004.13060 [Cs].