I think Invoke is competitive for now, but biggest advantage is an improved UX, and a large community with an ambitious roadmap focused more on enthusiasts/pros.
I’d give it a whirl and see where you end up preferring to do your SD projects :)
I legitimately think the answer is - - Experimental/novel SD features, Automatic typically has them first - Better UI/UX and exploration/gallery workflows, Invoke.
From a feature perspective, both have their own flavor of certain features.
Goal of Invoke is to eventually have prompt/concept library, a node-based workflow UI (with the ability to share techniques), etc.
It’s kind of a - switch if you want a better UX now; keep an eye on it if you want new workflow solutions long term.
- better inpainting
- a gallery to easily compare generations and easily regenerate images with small modifications
- responsive design --> works great on mobile, swipe left/right to switch between pictures in same generation and up/down to switch to another generation to compare
Here is the repo: https://github.com/leszekhanusz/diffusion-ui
Also if you don't have the hardware, you can also get images for free using the Stable Horde (https://stablehorde.net), a cluster of backends provided for free by volunteers.
You can test it here: https://diffusionui.com/b/stable_horde
The company whose code was stolen works closely with the man behind SD and the decision was made to merely ban him from the community instead of torpedo-ing the repo via DMCA.
It's not even a particularly large or interesting piece of code, the only reason why it's controvertial is because of the code is only necessary to use the NovelAI leaked models.
As you can see, that repo from 2 years ago even originates the "# attention, what we cannot get enough of" comment and is an exact 1:1 match to Automatics commit, while the one from NovelAI even has a small change in the if clause that Automatic doesn't have.
The bees knees would be being able to use automatic as a CLI or with some programmatic interface, because it is more feature rich. But I haven't seen anything that allows me to do that yet, so I'm stuck to its clunky UI or to use invoke-ai.