Anyone new to GAN art might not realize how big a deal this app is. The vast majority of art generators offload the calculations to their own curated model on their own server, which you have little to no control over. And computers previously needed some kind of Linux setup and a big stack of processing power to make anything, following a steep learning curve. Then this app comes along and freakin’ SOMEHOW takes full advantage of the recent iOS neural network architecture to keep generation totally local using models of your own choosing. On your -iPhone- no less. Including optimizers like LoRA, upscaling, inversion embedding, and a bunch of stuff you likely didn’t even know existed before. You can even train your own modules ala DreamBooth to personalize it for whatever you’re making!As further proof of miracles, it generates crazy fast for having it done on a calculator-sized smartphone, and has an interface that is tightly designed and responsive. Not to mention in-app quick downloads of a staggering number of models/modules alongside an easy option to upload your own. You won’t be able to make Van Goghs with 4 keywords and 0 AI experience, but having full control means you'll actually bother to learn and take advantage of a capability that was until very recently limited to GPU-stacked server farms. If you’re someone who dipped enough of a toe into AI art that you want to take the training wheels off and see just how much can be tinkered with, this app is peerless.

Update: Well , it’s happened. Liuliu has found a way to make this work on my phone! And it works well! My first attempt led to a hard crash and restart, but everything since has been smooth. And, surprisingly, I don’t seem to have any resolution limitations. I can generate 512x512 images just fine, though they take about five minutes at 30 steps. And one image will guzzle about 7% of my somewhat degraded battery life. But I can run Stable Diffusion 2.1 with negative prompts on my phone for free! It’s such a powerful tool for visualization. Thanks to the developer for all the hard work! I can’t find a way to generate at 768x768 with the native 768v model — I know I’m being greedy, but if it’s possible, please try to find a way!Original review: I absolutely support what the developer is doing here, as Stable Diffusion is a revolution in digital art tools and it is amazing that it can work on a mobile device. Unfortunately, I have not been able to get the app to work. Ever. The listing says my phone, an SE 2, is compatible, but the app crashes about 1/3 to 1/2 of the way through generation on even the smallest resolution image (384x384). I’m sure it’s running out of memory, but even quitting all other apps and setting the number of steps to the minimum doesn’t help. I don’t know what else to do. If there is a way to get lower-memory devices to work, the app should suggest the appropriate settings. And if not, the compatibility section of the app listing should be adjusted to better reflect what is actually possible. Dev, please help!

Specifically, on a 4K screen, the right sidebar showing the history shows images at low res scaled up so they appear pixelated compared to previewing the image you click on in the main middle column. Also, on such large screens it would be PERFECT if the grid based layout just shows up in the right sidebar where the history is instead of popping up in a tiny modal screen. I have a magic mouse and for some reason it doesn't scroll with inertia most of the time in this app. It could be the mouse, just that's part of the frustration of the current UI. Another is the inability to move through the history using the arrow quickly. The inability to right click on a history image and do more than just delete it. Would love to see updates to the UI to address these concerns and make it much much much easier to browse thousands of generated images quickly and more easily. There should be a way to end the current sequence of generations, not just skip 1 out of 100. If there's already a way, it's not obvious or intuitive. Also, can we please have an option to generate more than 100? Sometimes I like to leave it running all night. Also, can we please view and manage generated photos even while the app is generating new photos? That'd be great. I'll probably continue to edit this review as I think of new things, so check back often haha. Thanks so much for considering how you will update the UI soon!

I want to preface this by saying that this is ADVANCED AI generation software. I've seen a lot of 1 star reviews of this app by people who don't know what they're doing and are expecting to just type a prompt and get exactly what they want with no/minimal additional tweaking/setup. If that's what you want, use something like Midjourney. However, this software is virtually unparalleled in ease-of-setup and quality for using Stable Diffusion, you just have to do a little bit of learning and then it works virtually flawlessly. The app has support for LoRAs, textual inversion, and additive models. There are also plenty of quality-of-life features (like giving the option to delete models from your downloads folder once you upload to Draw Things.) There is also a wealth of settings to tweak as well as the option to clear the temporary cache to free up space. Underneath every setting is a well-writen and easy-to-read explanation detailing exactly what it does and what you might prefer to use. Massive respect and many thanks to Liu Liu for providing this software completely for free without any in-app purchases or subscriptions, you're brilliant. It's clear a lot of effort and care went into the user experience.