Offline Standalone Mac Installer for Stable Diffusion Web UI (unofficial) v1.1.1 Release Candidate #19
Replies: 31 comments 57 replies
-
|
Just given this a test drive on my 32GB M1 Max. |
Beta Was this translation helpful? Give feedback.
-
|
This is good news. In the previous version, the SD2.1 version produced an error. I don’t know why. |
Beta Was this translation helpful? Give feedback.
-
|
Not entirely sure, but I think macOS 13.3.1 messed up something with the generation on Intel Mac because now I can't get the correct images at all with any sampler. I've tried both your fork and the original one, each picture looks like green/brown/grey mush. But on a bright side - generations times now on par with ROCm implementations on Linux or even slighly better (at least for my RX6600XT), even if I can't get any image from it :^) |
Beta Was this translation helpful? Give feedback.
-
|
@p2mmao What do you get if you don't change from the defaults? Same as @Vitkarus? |
Beta Was this translation helpful? Give feedback.
-
yes |
Beta Was this translation helpful? Give feedback.
-
|
btw, the speed is 15% lower on 13.4 RC3 compared to 13.2.1 on my platform |
Beta Was this translation helpful? Give feedback.
-
|
@p2mmao Have you switched to the 13.4 full release yet? Wondering if it improves any on the slow down you had with the RC. |
Beta Was this translation helpful? Give feedback.
-
no, I've upgraded to 13.5 beta1,the speed returned to normal, also the GPU utilization, which never passed 50% |
Beta Was this translation helpful? Give feedback.
-
|
Great, I failed to upgrade Pytorch 2.0 , And Then used it for this installation package. The upgrade was successful! |
Beta Was this translation helpful? Give feedback.
-
|
Actually with this fork that setting should be set to |
Beta Was this translation helpful? Give feedback.
-
|
Okay, I set it to automatic and it runs well. |
Beta Was this translation helpful? Give feedback.
-
|
This build is running at essentially the same speed for me as the previous one, a bit faster on the initial pass and a bit slower on the Hires fix pass. But I don't see tomesd anywhere. Is it somehow "internalized", or will we see fresh speed gains when tomesd is formally merged into the base Automatic1111 and then flows into these builds? I have tomesd merged manually into the official Automatic1111 version I run, but this experimental build is still about 5% faster than my souped up official version. I hit the NaN error on some runs. I added the --disable-nan-check argument, but I get infrequent black images, so there is a real NaN issue going. In the past, for me, it was always a false warning. I've only used one model so far, a Frankenstein merge I did myself a ways back, so it may be the model that is at fault. I'll try with other models in the next day or two. Beyond that, this build seems solid, but I had no problems with the prior build as well. |
Beta Was this translation helpful? Give feedback.
-
|
I have the same problem with the NaN error in 1.1.1... (Mac Mini M1 8GB) At first it also happened in the 20230416 build, I always had to set "cross attention layer to float32, "automatic" would work sometimes, but mostly not... but after I played around with the ToMe Setting and disabled "Apply only to high-res fix pass", the crashes were gone and my machine ran like a beast ( 2.2s/it instead of 3.5s/it)...! Now the error is back and I miss my beloved ToMe settings... I'm not 100% sure this is the culprit, but I guess... Luckily I did a clean install and still have the older version... With these settings I had a blast (with 20230416):
|
Beta Was this translation helpful? Give feedback.
-
|
Thank you very much for the app! I couldn't use official Automatic1111 before due to high memory pressure and old version of PyTorch (for some reason couldn't update it). And your installer solved all the issues and now image on macbook air M1 with 8GB RAM (on Euler A with 20 steps) generates in 45-60 seconds, even faster then it should be according to guide in official Automatic1111 repo. So thank you very much again for the effort) |
Beta Was this translation helpful? Give feedback.
-
|
And here it is working. Loaded up and opened a browser window. The site-packages folder has 323 items, 98% are folders. |
Beta Was this translation helpful? Give feedback.
-
|
I'm off to dinner. Here is a link to the other app. If you go that route, leave a message here and I'll point you to models and a friendly user support group. If you stick with Automati1111, also let me know if you're able to get it running or not. I'm baffled by why you are not getting a complete install, so maybe we can do it step by step at the same time. https://github.com/godly-devotion/MochiDiffusion |
Beta Was this translation helpful? Give feedback.
-
|
I got it to work! I decided to update the computer's software to the latest version and started the download process over + followed your instructions.. now that I have it up do you have any resources for Automatic 1111 you'd recommend? Thank you again you're the best!! |
Beta Was this translation helpful? Give feedback.
-
|
The best, and pretty much only source for base models (checkpoints) is Civitai at: https://civitai.com/ Where there is a choice of versions for a given model, use the FP16 ones which usually list as 1.99 to 2.22 GB each. The ones that are double those sizes are FP32, and on a Mac, it doesn't do a lot of the math at FP16, because memory is limited, so the FP32 size models are a waste of space and download time. Besides checkpoints, you will also find LoRAs at Civitai. These are smaller models, 200 to 500 MB each, that add a specific visual look/style, or a famous person or character, or some sexual fetish, or some other niche thing. You use them together with a base model to get the special thing they provide from a base model that couldn't do it otherwise. There are also things called ControlNets that let you transfer some aspect from a starting image or a black and white simple drawing over to a generated full image. One ControlNet finds and transfers outlines of shapes, another finds and transfers poses of human bodies, another called Scribble takes a quick scribbled sketch idea and turns it into a detailed image based on your text prompt. There are other special features that get added by downloading extensions, like removing backgrounds from a starting image and then you can prompt the replacement background with text. They add new features to Automatic1111 monthly and they work their way into extensions or new releases of the download you installed today. Civitai is the place to see what all these things do. You can filter searches if you poke around the settings at the top right of a page. You can also find some tutorials there. New things get added at Civitai daily. Like 100 new models and LoRAs a day. If you are interested in erotic or NSFW stuff, you need to adjust your profile settings to let that all be visible. Same for violent images. You tell it what you do or don't want to see. Be patient with it all, as you were getting it installed. Automatic1111 is a very, very big world, so it takes time to learn it. I've seen some online books, but they get out of date as son as they appear, so you'll do just as well Googling things. If you want info on something specific, or get stuck on something, email me at [email protected] and I'll try to help. I'm retired. I have the time. This stuff is a hobby. I fool with some aspect of it, mostly technical, less so creative, every day. I'm glad you got it running! |
Beta Was this translation helpful? Give feedback.
-
|
ControlNet Info They let you instruct Stable Diffusion visually, instead of only with words. |
Beta Was this translation helpful? Give feedback.
-
|
Token merging ratio hr: 0.5. Previously, this parameter could be set, but is it not available in the new version? Feeling like the same seed map is a bit slow, right? |
Beta Was this translation helpful? Give feedback.
-
|
Thank you so much for this package. I'm getting a weird bug with this new version where performance quickly degrades as you continue to generate images. I'm on an M2 Mac Mini running Ventura 13.3.1(a) These are all the same parameters, with the only difference being small changes (<30px) in the img height. A 5th run ran even slower and I had to force quit Terminal.
|
Beta Was this translation helpful? Give feedback.
-
|
Playing around with it more, it's fine until I trigger it by doing something else when generation is starting (literally just web browsing). And once the performance is gone, it never comes back unless you restart everything. |
Beta Was this translation helpful? Give feedback.
-
|
Awesome. From not being able to use it (taking over 4 minutes to generate an image) to 30-50 seconds. Apple M1 Pro. Ty |
Beta Was this translation helpful? Give feedback.
-
|
Working flawlessly here. Significantly fewer NaN errors. Controlnet 1.1 running perfectly. And quickly. Mac air M2 16gb. |
Beta Was this translation helpful? Give feedback.
-
|
I'm unable to install deforum, I can download the extension and shows installed but the tab to select is not showing. Is there a fix for this? I haven't tried any other extensions yet. added a second screenshot of what shows when launching if it helps |
Beta Was this translation helpful? Give feedback.
-
|
same trouble |
Beta Was this translation helpful? Give feedback.
-
|
same for Roop |
Beta Was this translation helpful? Give feedback.
-
|
any news regarding extensions..? |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
waiting on roop @brkirch :) getting errors too |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for release! Tested on Mac M2, Ventura 13.3.1 |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the update! Crazy speed but NaN errors still show up randomly :( |
Beta Was this translation helpful? Give feedback.
-
|
sure~~ |
Beta Was this translation helpful? Give feedback.
-
|
controlnet 1.1 reference pre process can't work, why? |
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
怎么操作？ |
Beta Was this translation helpful? Give feedback.
-
|
I received that error on launch after renaming the original install folder. I had to name it back to what it was. Any chance you moved or renamed the folder after you installed the app? |
Beta Was this translation helpful? Give feedback.
-
|
we can draw in MacOS 14 sonoma
|
Beta Was this translation helpful? Give feedback.
-
|
On OSX 12.4 I get RuntimeError: "upsample_nearest2d_channels_last" not implemented for 'Half' FYI for those out there who wanted to try it on Monterey |
Beta Was this translation helpful? Give feedback.
-
|
Try adding |
Beta Was this translation helpful? Give feedback.
-
|
how to avoid this..: |
Beta Was this translation helpful? Give feedback.
-
|
You probably need to paste in more lines from the error message here so someone can see what |
Beta Was this translation helpful? Give feedback.
-
|
you are right... i am trying to install some extensions, unfortunately all of them are not visible after installation... in the terminal is pointed out several missing modules..: Launching Web UI with arguments: --autolaunch --skip-torch-cuda-test --skip-install --no-download-sd-model --no-half-vae --upcast-sampling --opt-sub-quad-attention --use-cpu interrogate --no-half 2023-06-25 17:14:40,236 - ControlNet - INFO - ControlNet v1.1.227 Error loading script: text2vid.py |
Beta Was this translation helpful? Give feedback.
-
|
I've seen a number of posts in the past week or two about problems with getting deforum and some other extensions to download, load, and/or run properly. I don't think anyone has figured out what is going on yet. Aside from that, at the very top it is saying that you need macOS 13+. That is real requirement. Support for important parts of stable diffusion pipelines only started with macOS 13.0 and were improved in 13.1. What version is your OS? The warning about xformers can be ignored. All Mac users get that warning. xformers does not run on a Mac, but Automatic1111 can run just fine without it. |
Beta Was this translation helpful? Give feedback.
-
|
i still use os 12.6 with the 13 system there are e.g. massive problems with external sandisk ssd hard drives - up to total data loss... = i can only upgrade to the 13 when the SSD problem is fixed... |
Beta Was this translation helpful? Give feedback.
-
|
I understand completely. But you may not be able to run many of the Stable Diffusion implementations. DiffusionBee says it runs on macOS 12.5.1 or higher. https://diffusionbee.com/ |
Beta Was this translation helpful? Give feedback.
-
|
I will soon see if the problem with the hard drives has already been solved... |
Beta Was this translation helpful? Give feedback.
-
|
FWIW, I have been using 3 Samsung T7 external SSDs with my M1 Macbook Pro, including with macOS 13 since it became available. I have not had any problems, in fact I boot the macOS 14 beta off of one, rather than risking my regular install with a beta. On the other hand, I have had some problems with large name brand SD Cards that caused large data losses. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the works @brkirch , I think this project is very helpful for mac users. But I can see this project is 14 commits ahead, 730 commits behind AUTOMATIC1111:master. So my question is, are we going to have new version of this project? |
Beta Was this translation helpful? Give feedback.
-
|
Thank so much for this build! The Speed is perfect. |
Beta Was this translation helpful? Give feedback.
-
|
I saw a guide on how to activate/use xformers on the M1 mac via Rosetta, though I haven't given it a shot since I don't want to mess up this working SD for my Mac M1, anyways has anyone tried it? Interested on the boost that xformers can give on the mac m1 compared to the windows version --xformers. Thoughts? Thanks guys |
Beta Was this translation helpful? Give feedback.
-
|
I'e been running SD for a while on my M1 but thought I'd give this a go for better performance. Install and getting the following error when running: ERROR: python3-venv is not installed, aborting... Any ideas? |
Beta Was this translation helpful? Give feedback.
-
|
Did you install from one of the tar.xz file in Resources? If you installed by cloning this repo, it will not work. The repo wants you to use one of the app bundles available in Resources. |
Beta Was this translation helpful? Give feedback.
-
|
Yea, downloaded the latest version, ran the install App. |
Beta Was this translation helpful? Give feedback.
-
Beta Was this translation helpful? Give feedback.
-
|
Hmmm. Don't know then. I once had issues that seemed to come from renaming the base folder, and another from dragging the start-up script applet to a different place, rather than using an alias to it from that different place. Did you get the sense that there was any first-run setting up going on, or did it go immediately into the error message? A first run can take a few minutes. |
Beta Was this translation helpful? Give feedback.
-
|
Anyone who was able to make roop work on this version? 👌🏻 |
Beta Was this translation helpful? Give feedback.
-
|
no help? lol |
Beta Was this translation helpful? Give feedback.
-
|
Silence usually means “no” in these sorts of places. If someone gets something to work, they are eager to tell the world, and right away. |
Beta Was this translation helpful? Give feedback.
-
|
any luck with ADetailer as well? giving me bunch of errors starting the "RICH" module not found., tried to do a manual transfer from my windows laptop, unfortunately the webui itself won't launch after transferring the "RICH" folder. So eager to try that one. Thanks man |
Beta Was this translation helpful? Give feedback.
-
Both roop and ADetailer work fine. |
Beta Was this translation helpful? Give feedback.
-
|
was able to figure this out already Thanks. what about the Optimization for the Token Merging? |
Beta Was this translation helpful? Give feedback.
-
|
hi, i would like to use img2img with batch.... many THX, Lassinger |
Beta Was this translation helpful? Give feedback.
and others
-
Changelog
This discussion was created from the release Offline Standalone Mac Installer for Stable Diffusion Web UI (unofficial) v1.1.1 Release Candidate.
Beta Was this translation helpful? Give feedback.
All reactions