this post was submitted on 29 Jul 2023
35 points (100.0% liked)
Stable Diffusion
4310 readers
5 users here now
Discuss matters related to our favourite AI Art generation technology
Also see
Other communities
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I had hoped that would sell a few people on it! I agree entirely on the motivation - I was able to test it on work machines without even needing to log out of an unprivileged user thanks to the portable install working nicely for me. MPS is of course slower than an equivalent CUDA device, but I was able to ensure the entire E2E test plan passed on Mac, including all ControlNets, inpainting, schedulers, upscaling, etc.
If you want SDXL on Mac, your mileage will definitely vary. I ran out of memory while loading the checkpoint on my M1 Pro 12GB, it may have been able to work if I allotted it a dangerously large amount of memory, but I could also have crashed the machine and I don't feel like bothering with that. In theory there's nothing stopping it from working, you just might need an M2 Max to get it off the ground.
Please let me know if you encounter any unforeseen issues!
So, here's my findings. Easy install, portable enough (having to specify a bunch of folders and manually creating them could be better) at first sight the interface is nice. And that's really where I stop, because it took... what, I think it was over 5 minutes to initialize the first render? All to error out (gracefully! Kudos for that) because it was out of memory. I couldn't find anything else to close, so basically on the M2 with 8gb it won't run. It was 512x512 SD 1 model. https://apps.apple.com/app/id6444050820 works on all sorts of Apple devices, is free, and kept very updated... I mention it because it's fast (14 seconds 512x512 20 steps V1 on M2), can do SDXL and refiner even with 8gb (once, I doubt many will do it a second time, but a couple of minutes for 1024x1024 20 is still "doable"). I'll stick with that on Apple stuff :)
I had hopes to try it on the Steam Deck but I saw no mention of AMD at all. Still! I'm probably going to try the tensorRT stuff on Windows, my 3060 should do it and I don't know how to do it with Automatic1111 XD
I can't thank you enough for linking that!!!
It made me realize that there must have been a way to effectively downcast without getting NaN's. There's just no way that app could work on the devices it does with SDXL without having figured it out, so I scoured the web for references and dug in to figure it out. I'm happy to say I got it working on my M1 Pro! That also means memory usage is cut down by about a third, and speed is up by about 50% on Mac in general thanks to being able to work in half-precision instead of full.
I was able to do the same 512x512 20 steps in 17 seconds using a fine-tuned model (Realistic Vision 5.) SDXL took it's sweet time coming in at almost 3 minutes, so yeah probably not my usual workflow, but SDXL isn't even in my usual workflow on my 3090 Ti Windows/Ubuntu hybrid machine. I still use TensorRT and fine-tuned SD 1.5 models - 512x512 is roughly 3 seconds on that, but the beautiful part is when doing a 2000-iteration upscale and TensorRT caps out at ~30 it/s on Windows or ~40 it/s on Linux.
I have a little bit more testing to do for this, but I'm going to be releasing a 0.2.1 build in the next couple days. I would love it if you would give it another shot - I'll send you a message with a link, if that's okay with you!
With respect to AMD - that's a complicated answer, I'm trying to work with some AMD users to test out the combination of dependencies that will work for them. I'm not sure if anyone has managed to successfully use the GPU for AI on the steam deck, but I do know officially ROCm is unsupported and will be for the foreseeable future on the deck.. I've seen people successfully use Stable Diffusion with CPU inference on it, which Enfugue will allow - but those same people reported it took half an hour to generate a single image on the deck, so I'm not sure it's worth trying.
Really glad that could help! Since I've got your attention, I couldn't get TensorRT to work on Windows. At least 50% chance I didn't install it properly, BUT at the same time your gui was showing my 1650 instead of the 3060. After looking for some setting about Cuda devices and finding none I gave up. Generation times and usage pointed clearly at a normal 3060 task, even if the gui had the temperature for the 1650.
But anyway! One thing I'd like to ask is (now that there's a viable way to use it on my Mini) an option to allow other computers to access it, and better yet the API like in Automatic1111. Like that I could do some kind of LLM on the 3060 (I like Pygmalion 6b) and stable diffusion on the Mac.
All that aside, thanks for making a viable alternative to Draw Things. As much as I like it and the interface, choice is always good... and yours has the potential to be usable in remote :D
I'm back! 0.2.1 is now released, which defaults MacOS to half-precision. It also includes SDXL LoRA and ControlNet support, which I did get working on my Mac. :) It's available at https://github.com/painebenjamin/app.enfugue.ai/releases/tag/0.2.1.
As for API - there always was one! It was just never documented until now. There's still a few endpoints left to document, but the big ones are covered. Documentation is at https://github.com/painebenjamin/app.enfugue.ai/wiki/JSON-API.
So, feedback. To begin with, it works! That's a massive improvement and allowed me to actually try it. Civitai.com downloading works quite nicely and... the generation is kinda slow. Slower than my iPhone 13 pro with Draw Things, a minute give or take 10 seconds. Poor phone crunches the same model in 30 something seconds.
Don't get me wrong, I appreciate it works to begin with, it's also easy to setup, but there's a fair amount of performance left on the table. Now depending on how much work there's to do it might make sense to chase further performance, but that's something only you can decide :D
You're the best, thanks so much for trying it and getting it working!
I don't think it's ever not worth chasing improved performance, so I'm definitely going to continue looking for optimizations. While cannibalizing the code for Comfy and A1111, I saw a lot (and I mean a lot) of shortcuts being made over the official Stability code release that improves performance in specific situations. I'm going to try and see how I can leverage some of those shortcuts into options for the user to tune to their hardware.
This latest release has attracted some more developer attention (and also some inquiries from hosting providers about offering Enfugue in the cloud!) I'm hoping that some of the authors of those improvements find their way to the Enfugue repository and perhaps are inspired to contribute.
With that being said, TensorRT will definitely knock your socks off in terms of speed if you haven't used it before, if you've got the hardware for it. I'd be happy to troubleshoot whatever went wrong with your Windows install - there should be up to three
enfugue-engine.logfiles in your~/.cache/directory that will have more information about what went wrong, if you'd like to share them here (or we can start a GitHub thread if you have that.)Thank you again for all your help!
Now knowing where to look, I did some fixing by myself! Main issue is that I had CUDA 10 and 12, no 11. Then after going insane about that tiny difference... I landed on something I lack the knowledge to decipher: "PyInstallerImportError: Failed to load dynlib/dll 'C:\Program Files\NVIDIA GPU Computing Toolkit\TensorRT-8.6.1.6\lib\nvinfer_plugin.dll'. Most likely this dynlib/dll was not found when the application was frozen."
All I can say is that the file is there.
Hey! I am able to reproduce that error by using the CUDA 12 version of TensorRT.
PyInstallerImportError: Failed to load dynlib/dll 'C:\\TensorRT-8.6.1.6\\lib\\nvinfer_plugin.dll'. Most likely this dynlib/dll was not found when the application was frozen.Please make sure you downloaded the top file here, not the bottom.
I was able to modify my PATH and point to the right TensorRT, then restart the server, and it worked for me (no machine restart needed.)
Please let me know if that works for you :)
My request is dumb, the UI is glitching a little but hot damn 12 iterations per second! Impressive.
YOU GOT IT WORKING?
You are the first person to stick through to the end and do it. Seriously. Thank you so much for confirming that it works on some machine besides mine and monster servers in the cloud.
The configuration is obviously a pain point, but we're running along the cutting edge with TensorRT on Windows at all. I'm hoping Nvidia makes it easier soon, or at least relaxes the license so I'm not running afoul if I redistribute required dll's (for comparison, Nvidia publishes TensorRT binary libraries for Linux directly on pip, no license required.)
It's also a pain that 11.7 is the best CUDA version for Stable Diffusion with TensorRT. I couldn't even get 11.8, 12.0 or 12.1 to work at all on Windows with TensorRT (they work fine on their own.) On Linux, they would work, but would at best give me the same speed as regular GPU inference, and at worst would be slower, completely defeating the point.
Not going to lie, I almost gave up a few times. But I can also be stubborn... anyway since this is apparently the first confirmation it works, it's probably be helpful if I mention that it's a 12gb 3060. :)