Adobe’s Photoshop has been the gold standard for raster image editors for decades, and that trend has continued even with AI-powered apps shaking things up in the productivity ecosystem. After all, Adobe integrated a bunch of useful AI utilities into the uber-popular image editor, with background removal, smart subject selection, and generative fill being absolute game-changers for touching up images.

That said, rival tools have come a long way over the last couple of years. The best part? If you despise cloud platforms that charge monetary subscriptions in exchange for their clanker services as much as I do, you can host image generation models on a local GPU and use AI features on FOSS apps. In fact, I’ve been tinkering with an AI-heavy Krita instance for the last couple of days, and it’s mind-boggling how well this free setup works for my image-editing needs.

I’ve grown disillusioned with Adobe tools over the years

Pricey subscriptions and Adobe’s anti-consumer measures are deal-breakers

Despite using Photoshop like a madman as a starry-eyed teenager, I’ve started hating Adobe’s app ecosystem as of late. The removal of a perpetual license and the slow transition to a subscription model was a massive letdown for my broke self, especially since CC’s monthly fees aren’t cheap by any means. The AI features also require generative credits to run, and although the rate limits aren’t as low as the API caps on certain coding tools, it’s not hard to hit that number and end up with painfully slow generation speeds if you rerun the AI tools multiple times during a Photoshop session.

Plus, I’m not fond of the anti-consumer antics Adobe has repeatedly pulled over the years. In particular, the fact that Adobe attempted to modify the terms and conditions to grant the company full ownership over its userbase’s creations was the straw that broke the camel’s back, even though it later went back on its word after receiving massive backlash from the artistic community. Toss in the lack of privacy on queries and generative content sent to Adobe’s cloud, and you can see why I decided to break free from Photoshop.

Krita and its AI Diffusion plugin can trade blows with Adobe’s generative features

It’s easy to set up and works better than you’d expect

When I set out to find a completely local AI-powered image editor, I initially gravitated towards GIMP and all the community-created plugins on GitHub. Unfortunately, most of the highly rated ones hadn’t been updated in a few years, and I encountered one failure after another trying to set up the newer repos, regardless of whether I used Windows or Linux.

That’s when I ran into Krita AI Diffusion, a game-changing plugin that replicates most of Photoshop’s generative tools while running everything locally via ComfyUI workflows. Setting it up was a piece of cake, as the plugin includes an automated installer for ComfyUI, models, and image generation workflows, so I didn’t have to waste hours troubleshooting random Python compatibility errors or scurrying around looking for discontinued repos.

On the features front, Krita AI Diffusion supports both inpainting and outpainting, meaning it’s as capable of editing stuff inside the image as it is at stretching the background and extending elements beyond the initial canvas. Truth be told, I wasn’t very sure about its accuracy, but a few minutes of playing with Flux 2.0 made me change my mind.

I’ve tried generating multiple abstract objects, ranging from rings for my fingers to additional wires for images highlighting home lab hardware, and the results were really impressive. The only situation where the plugin fell short was when I tried to generate an image of a Raspberry Pi, as its SBC components were completely off. But for generative infill tasks involving simple objects, especially those concentrated on a specific region of a larger image, it’s a perfectly viable alternative to Photoshop.

Likewise, the plugin is great at expanding canvases, and it can even upscale images past 8K without introducing visual artifacts. There’s even a live painting facility, which uses simple brush strokes and text descriptions to modify the canvas into fairly precise outputs. That's pretty handy when I want to add new objects to an existing image. I haven’t experimented too much with line art or illustration generation, as I plan to use this setup exclusively to edit images. But it’s pretty good at turning simple drawings into proper illustrations, provided I add a proper description and regenerate the layers a few times.

The Krita Vision Tools plugin takes care of my auto-selection and background removal needs

Besides Krita’s AI Diffusion plugin, the Vision Tools package is just as necessary for this AI-powered pipeline. That’s because it adds smart selection and background removal to the FOSS app – and both facilities are capable of harnessing my GPU. The Select Segment from Box tool works well across dozens of images with radically different subjects, though I had to set it to Precise mode to avoid the AI model including unnecessary objects in the selection.

Then there’s the background removal feature that’s hidden deep in the Filter tab. I’ve used it in tandem with the Select Segment from Point tool, and although I had slight overlaps with redundant elements when attempting to get rid of the background on landscapes, I don’t have too many complaints about this plugin, either. Considering that I’ve got everything running on an RTX 3080 Ti with 12GB of VRAM, it’s surprising how well Krita’s plugins work for local image generation tasks.