Now that Apple has added native CoreML support for Stable Diffusion in Ventura, it'd be amazing to see Stable Diffusion in Pixelmator Pro. Namely:
Text-to-Image generation, with the ability to determine the dimensions, CFG, which samplers to use in the generation, and prompt weighting.
Image-to-Image generation, with the same abilities above, and additionally the ability to inpaint/outpaint on separate layers, while also being able to infer from other layers what the image-to-image generation would be.
Model (ckpt), CLIP embeddings, hypernetwork, and VAE support as well. The ability to allow users to download or use their own models, embeddings, and other things listed above would be highly helpful in allowing us to put our own stamp and style on what we're creating in Pixelmator.
An example of how this can work (and is already working in Photoshop) can be seen here: https://www.reddit.com/r/StableDiffusio ... hop_stable
I think that, especially now that Pixelmator Pro has video editing capabilities, adding in native Stable Diffusion generation would make it the creme de la creme. There is a plugin for SD already in Krita, and Adobe has announced that they will be supporting it (eventually), though there are a few third party plugins already made for it (as given above).
I know there's a lot of angst and fear with this new AI tech. Without devolving into all that, I think that it's important for Pixelmator as a tech product (as well as art product) to keep up with the times. It's becoming clearer every day that AI generation isn't going away soon, and in all likelihood, it will be an integral part of the future of visual arts and graphic design, just as photo/bitmap editing software has become.
Stable Diffusion in Pixelmator
2022-12-21 18:49:02
2022-12-22 14:19:03
Yeah, this would be super cool.
2022-12-22 18:56:54
I think so, too!
Oh, and since Pixelmator itself is big on UI/UX, here's something that is really neat (that I forgot to mention above): https://www.youtube.com/watch?v=RwVGDGc6-3o
It's an app called InvokeAI that allows users to generate on an infinite canvas using Stable Diffusion (they call it their unified canvas). The unified canvas is great, and even allows you to mask certain parts of an image so as to only generate on the mask of the image (which would be a neat feature as well to add). Overall, its got a great UI and UX, and hopefully gives another look at what working with Stable Diffusion in Pixelmator could be like if it's all integrated.
Oh, and since Pixelmator itself is big on UI/UX, here's something that is really neat (that I forgot to mention above): https://www.youtube.com/watch?v=RwVGDGc6-3o
It's an app called InvokeAI that allows users to generate on an infinite canvas using Stable Diffusion (they call it their unified canvas). The unified canvas is great, and even allows you to mask certain parts of an image so as to only generate on the mask of the image (which would be a neat feature as well to add). Overall, its got a great UI and UX, and hopefully gives another look at what working with Stable Diffusion in Pixelmator could be like if it's all integrated.