Onirica: generate and edit images with AI, locally

The app itself is not 15 gb, but the download includes also the AI model as an asset, which is why the total dimension will be around 15 gb. It's quite similar to when you download videogames, basically.

Thank you for the comment. It's the first time I'm sharing this app, so I could communicate it better, for sure. And comments like this help a lot. For your question, if I release a functional update to the app itself, like exporting images in other formats, there won't be another 15 gb download. If I change the image model itself, then the new model will be downloaded to replace the old one. I hope it's clear enough!

Thanks! Unfortunately, there's no way to do it in the App Store.

Thanks, let me know what you think! And if you like it, a review will be much appreciated :)

Good question. I think Draw Things is excellent, and probably the most powerful option out there if you want deep control: LoRA training, ControlNet, custom models, scripting. The tradeoff is that power comes with a learning curve. Even people who like it describe it as intimidating at first, there's a lot of settings to understand, models to find and manage separately, and the general advice is not to touch things you don't understand yet. It's a power user tool first. Onirica is the opposite bet. Open it, type what you want, get an image. No model picking, no settings to learn first, editing and restoring photos work the same simple way as generating. If you want a deep toolbox, Draw Things is great and free. If you want something that just works with zero setup, that's what I built Onirica for.

Thank you! The model is not loaded in ram all at once, but it first loads the part that decodes the prompt, and then the part that generates the actual image. That's why it can run also on 8 GB macs, even though it'd be better to close most of other apps in that case. I've tried it on my M1 Macbook Air with 8 GB (which is basically the worst mac you can run it on) and the generation of one image took around 4 minutes. On my M4 Mac Mini with 16 GB it takes about one minute.

Thank you! Unfortunately I don't think so, since the model is managed via Apple Background Assets, so it's basically downloaded from Apple's servers to a sandboxed container on the internal disk. There's no option in that framework to point it at an external volume. Sure! I comfortably run it on my M4 Mac Mini with 16 GB, it takes around 1 minute to generate each image.

Thanks. I've been able to run it on a M1 Macbook Air with 8GB ram, which is the worst you can run it on. Of course in such case it's better to close most of other apps. At peak, from my tests, it uses about 4/5 GB of ram. So, if you have 16 gb or more, it will be better in terms of running it while also running other apps.

Thanks!

Thank you! :)