Post ASWYAdpFaQd2Vdv4LI by InceptionState@poa.st
(DIR) More posts by InceptionState@poa.st
(DIR) Post #ASWXpdRpnOGxFywz56 by graf@poa.st
2023-02-10T03:31:26.346466Z
0 likes, 0 repeats
@Yeet give link to this setup, i want to make my 3070ti hurt
(DIR) Post #ASWY0BCt8Sucr0zHaC by InceptionState@poa.st
2023-02-10T03:33:22.170424Z
1 likes, 0 repeats
@graf @Yeet github.com/AUTOMATIC1111/stable-diffusion-webuiSeems to be the most popular front-end for most people, setup is pretty straightforward. Just takes a bit to download the dependencies and models.
(DIR) Post #ASWY0sXc1I2VFk0wM4 by graf@poa.st
2023-02-10T03:33:28.543200Z
0 likes, 0 repeats
@Yeet oh i thought you were using the self hosted thing. @grey you got a guide?
(DIR) Post #ASWY73dBhV4TNwZIie by grey@poa.st
2023-02-10T03:34:36.436579Z
0 likes, 0 repeats
@graf @Yeet novelai? i was trying to get stable diffusion working unsuccessfully, haven't played with novelai
(DIR) Post #ASWYAdpFaQd2Vdv4LI by InceptionState@poa.st
2023-02-10T03:35:15.605117Z
0 likes, 0 repeats
@Yeet @graf If you can host a Matrix server and one of the most popular Fedi instances, this is absolutely trivial in comparison
(DIR) Post #ASWYJrqmoPjIc32N9M by InceptionState@poa.st
2023-02-10T03:36:54.760432Z
1 likes, 0 repeats
@Yeet @grey @graf here you go github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-NVidia-GPUsYou download the NovelAI model separately and just drop in the models directory.
(DIR) Post #ASWYitDmqshm6Wj8pk by grey@poa.st
2023-02-10T03:41:26.647052Z
3 likes, 0 repeats
@InceptionState @graf @Yeet i'll try this, i fucking hate managing python installations with a passion which all this ai shit seems to use. so this script seems to handle that so i'll see how it goes.
(DIR) Post #ASWYrhkyqXgF9aSMNM by InceptionState@poa.st
2023-02-10T03:43:02.335272Z
0 likes, 0 repeats
@grey @graf @Yeet Yeah it handles all the python package management for you, which is nice. I only had to manually tweak something in the venv once, and that was after months of use. Just remember to do a git pull once in a while to keep things up to date.
(DIR) Post #ASWYz09kps4GFJWWkC by graf@poa.st
2023-02-10T03:44:20.526379Z
3 likes, 0 repeats
@InceptionState @grey @Yeet im running w10 on this machine right now with gentoo in WSL2 so I will try this lol
(DIR) Post #ASWZ5PYzHn8smXNmgy by grey@poa.st
2023-02-10T03:45:30.613735Z
1 likes, 0 repeats
@InceptionState @graf @Yeet lol and python still bitches, fuck i hate pythonWARNING: You are using pip version 22.0.4; however, version 23.0 is available.You should consider upgrading via the '/home/nigger/stable-diffusion-webui/venv/bin/python3 -m pip install --upgrade pip' command.
(DIR) Post #ASWZ7VdPgRvUBuS9dg by graf@poa.st
2023-02-10T03:45:52.519469Z
1 likes, 0 repeats
@grey @InceptionState @Yeet dont forget --user since gentoo cries if you dont
(DIR) Post #ASWZAem0ABf5P0yGbA by InceptionState@poa.st
2023-02-10T03:46:27.923910Z
0 likes, 0 repeats
@grey @graf @Yeet Just update pip and it will go away. It is annoying though.
(DIR) Post #ASWZDauoxkrEJ4Usc4 by grey@poa.st
2023-02-10T03:46:59.608575Z
1 likes, 0 repeats
@graf @InceptionState @Yeet this is being run by the sd installer in it's own directory, i'm not touching the pip command manually. i just don't get how python can be this bad at compatibility that so much shit breaks between versions that it needs this level of babysitting.
(DIR) Post #ASWZJGgp4goRJaeffE by graf@poa.st
2023-02-10T03:48:00.289395Z
3 likes, 0 repeats
@InceptionState @grey @Yeet negged by WSL2
(DIR) Post #ASWZLc3AczqS8tE480 by InceptionState@poa.st
2023-02-10T03:48:26.826376Z
1 likes, 0 repeats
@Yeet @graf AI: is this a bow?
(DIR) Post #ASWZXQ4YzBtuTBVVvU by keith@nightshift.social
2023-02-10T03:50:33.176620Z
1 likes, 0 repeats
rewrite windows in rust.
(DIR) Post #ASWZaUtVW1XuICnsi8 by graf@poa.st
2023-02-10T03:51:06.989577Z
2 likes, 0 repeats
@keith @Yeet @InceptionState @grey I might just rewrite windows on this NVMe with Gentoo
(DIR) Post #ASWZatFIydF5olBd1E by grey@poa.st
2023-02-10T03:51:12.175185Z
0 likes, 0 repeats
@InceptionState @graf @Yeet so does this mean i need to install a model or something alongside it? No module 'xformers'. Proceeding without it.No checkpoints found. When searching for checkpoints, looked at: - file /home/nigger/stable-diffusion-webui/model.ckpt - directory /home/nigger/stable-diffusion-webui/models/Stable-diffusionCan't run without a checkpoint. Find and place a .ckpt or .safetensors file into any of those locations. The program will exit.
(DIR) Post #ASWZeCqyX6BNUxO6hE by skylar@wolfgirl.bar
2023-02-10T03:51:43.540586Z
1 likes, 0 repeats
@graf @Yeet @InceptionState @grey it works fine in ordinary w10 without the gentoo
(DIR) Post #ASWZpe9xLjdnN8imDQ by InceptionState@poa.st
2023-02-10T03:53:52.568191Z
0 likes, 0 repeats
@grey @graf @Yeet It's complaining because it is using a slightly out of date version of pip, because that is what SD configured it to use when setting up the venv and pip has been updated in the meantime. It won't cause any issues, just present an annoying reminder to update. If you want to make the message go away, just update the venv's version of pip.
(DIR) Post #ASWa5TstJvTg8aI0x6 by InceptionState@poa.st
2023-02-10T03:56:44.243068Z
1 likes, 0 repeats
@grey @graf @Yeet The repo just installs the web ui, not the models. Eg. if you want SD 1.5, you can download it here:huggingface.co/runwayml/stable-diffusion-v1-5/tree/mainDownload v1-5-pruned.ckpt or v1-5-pruned.safetensors and drop it into the model folder in the web UI install directory.
(DIR) Post #ASWaCNfCqaqCs5B82i by grey@poa.st
2023-02-10T03:57:58.728610Z
1 likes, 0 repeats
@InceptionState @graf @Yeet isn't there a model focused on anime waifus?
(DIR) Post #ASWaIrO5FQ0QnOG8AK by InceptionState@poa.st
2023-02-10T03:59:09.302579Z
1 likes, 0 repeats
@grey @graf @Yeet Yes, loads of them. Go to civitai.com/, find one you like, download it, drop it into the model folder, and give it a spin! The top-left corner in the web UI has a dropdown menu to choose which model to use.
(DIR) Post #ASWadaWh8FQ0aDAeUS by grey@poa.st
2023-02-10T04:02:52.681565Z
0 likes, 0 repeats
@InceptionState @graf @Yeet nice, thanks, we'll see if this works. so theoretically after putting these files in that dir, i just run the webui script and connect to whatever port.
(DIR) Post #ASWax6xSm9n8fd07Ci by InceptionState@poa.st
2023-02-10T04:06:25.780311Z
1 likes, 0 repeats
@grey @graf @Yeet Yup, run it with eg. `./webui.sh --medvram --xformers`, connect with your browser, and then have fun messing with the settings. I'd recommend starting with 512x512 or 600x800 especially if you have a less powerful GPU, that way you can do batches of 3-4 without running out of memory.
(DIR) Post #ASWbA5hKlwCsWie6KG by grey@poa.st
2023-02-10T04:08:46.191734Z
1 likes, 0 repeats
@InceptionState @graf @Yeet
(DIR) Post #ASWbtfWkPGtkgFQYAi by grey@poa.st
2023-02-10T04:17:00.400184Z
0 likes, 0 repeats
@InceptionState @graf @Yeet lol, looks like i got cucked by my vram. will maybe need to not run 3 monitors while doing this i guess. torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 1.67 GiB (GPU 0; 7.79 GiB total capacity; 1.64 GiB already allocated; 1.69 GiB free; 3.33 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
(DIR) Post #ASWc8WXe63rJ9jvlwG by InceptionState@poa.st
2023-02-10T04:19:41.845053Z
0 likes, 0 repeats
@grey @graf @Yeet Yeah or do smaller batches and/or lower resolutions so it uses less VRAM at once. That's less efficient though for obvious reasons.
(DIR) Post #ASWcSUfEmrWt7n0JeK by graf@poa.st
2023-02-10T04:23:17.160003Z
1 likes, 0 repeats
@grey @InceptionState @Yeet I got my 3070ti set to highest memory use and 3 monitors, generating highest resolution image rn and the fans sure are loud
(DIR) Post #ASWd5jkKe6FbC96EsK by graf@poa.st
2023-02-10T04:30:22.673797Z
2 likes, 0 repeats
@grey @InceptionState @Yeet this is the first one my gpu made with very limited promptswhen i have some time to dedicate to this im gonna have fun i think
(DIR) Post #ASWdzM6ti92g6cAwka by grey@poa.st
2023-02-10T04:40:26.444073Z
0 likes, 0 repeats
@graf @InceptionState @Yeet what model are you using? i get weird looking hands and bodies still. but i only have a 2080 on here.
(DIR) Post #ASWe1Gy9nPl4jy099E by grey@poa.st
2023-02-10T04:40:47.411795Z
3 likes, 0 repeats
@InceptionState @graf @Yeet yeah, disabling two of the displays has it working. guessing there's alot of tweaking to do.
(DIR) Post #ASWe57GpnW77ooZQIq by graf@poa.st
2023-02-10T04:41:28.318502Z
0 likes, 0 repeats
@grey @InceptionState @Yeet that was this one just to test it was functioning civitai.com/models/1366/elldreths-lucid-mixill look more into models later
(DIR) Post #ASWe8NNYMbVse9Gdou by Ivan_Ivanovich@poa.st
2023-02-10T04:42:04.769449Z
1 likes, 0 repeats
@graf @grey @InceptionState @Yeet crypto mining rig turned AI anime girl generator
(DIR) Post #ASWeAUT4yeltDDrP7Y by grey@poa.st
2023-02-10T04:42:27.435157Z
0 likes, 0 repeats
@graf @InceptionState @Yeet i'll give that a shot, better than this anime one
(DIR) Post #ASWeDlmb3e0DmXYME4 by InceptionState@poa.st
2023-02-10T04:43:03.336424Z
0 likes, 0 repeats
@grey @graf @Yeet Handy thing - you can go to the PNG Info tab in the webui, drag and drop in an image, and it will show the metadata. Eg. in this case:> parameters> extremely high detail, beautiful female, elf girl, long ears, hetero chromia, red eye, green eye> Steps: 20, Sampler: Euler a, CFG scale: 7, Seed: 341552444, Size: 512x512, Model hash: 4b0eec4da8, Model: kenshi_01I have found kenshi a bit temperamental. It looks great when you get a good set of prompts, but it seems to need a lot of specificity.
(DIR) Post #ASWgiErhh7ZcZibAJ6 by grey@poa.st
2023-02-10T05:10:57.947388Z
2 likes, 0 repeats
@InceptionState @graf @Yeet it also seems to have a thing for making them all catgirls
(DIR) Post #ASWhaHHkucvLicYH0i by InceptionState@poa.st
2023-02-10T05:20:44.113796Z
0 likes, 0 repeats
@grey @graf @Yeet FYI you can dial back the steps quite a bit, I usually do 20 steps with DPM++ 2M Keras for the sampling method, or ~30 with Euler / Euler a. If you get something you really like you can always rerun it with the same seed and more steps.
(DIR) Post #ASWhlLSnibMtrWzmT2 by grey@poa.st
2023-02-10T05:22:43.806507Z
0 likes, 0 repeats
@InceptionState @graf @Yeet i tried rerunning with the same seed and prompt at a higher resolution and it gave me a different image? maybe it was the wrong seed
(DIR) Post #ASWhwsuVhhNEmWSVQO by InceptionState@poa.st
2023-02-10T05:24:49.307072Z
0 likes, 0 repeats
@grey @graf @Yeet I think that's normal, but in the img2img tab there are resizing options that should let you upscale a base image.
(DIR) Post #ASX8edv1Oxm9prDx5M by cyclonebee@poa.st
2023-02-10T08:40:36.215061Z
0 likes, 0 repeats
@graf @grey @InceptionState @Yeet Sir, "very limited" is a bit of an understatement
(DIR) Post #ASX8eeShNkiLWIyqZs by cyclonebee@poa.st
2023-02-10T08:54:16.505599Z
1 likes, 0 repeats
@graf @grey @InceptionState @Yeet - Check the LAION-5B for keywords you're going to prompt, that's what the "base" of stablediffusion is trained on. For Heterochromia for instance rom1504.github.io/clip-retrieval/?back=https%3A%2F%2Fknn.laion.ai&index=laion5B-H-14&useMclip=false&query=heterochromia you get pretty much medical reference pictures of realistic irises, but images for the animays such as "odd eyes red green (rozen maiden Suiseiseki)" seem to be also in the dataset. It's also closer to what you want as a style so I'd use that in the prompt over heterochromia.- weight your prompts. (prompt:value) gives more emphasis while [prompt:value] gives less emphasis. (big boobs:1.3) [fat thighs:0.3] is the same as [big boobs:-0.3] (fat thighs:0.7).- Check for LoRAs or hypernetworks for specific things like big booba, heterochromia and general animay- negative prompt is almost as important as the actual prompt- don't be afraid to go overly verbose, my own prompts are most often around 225 tokens long and negative prompts around 150
(DIR) Post #ASX8ef3v9MULNkOZay by cyclonebee@poa.st
2023-02-10T08:59:41.699062Z
0 likes, 0 repeats
@graf @grey @InceptionState @Yeet Quickly modding my current prompt with me model, adding your prompt and swapping heterochromatia for that animay reference I have no idea about (if it's a groomershit anime, don't at me because I wouldn't know) we start to get heterochromatia in the eyes. This time the RNG gave red-blue heterochromia but you get the drift
(DIR) Post #ASX8efbb89QX4C9T5U by cyclonebee@poa.st
2023-02-10T09:02:04.047707Z
0 likes, 0 repeats
@graf @grey @InceptionState @Yeet And note, this was done on a model aimed more at photorealistic/westernoid art, an anime model will be way better at those same tokens, weighed correctly
(DIR) Post #ASX8eg79EqfEe2ufGS by graf@poa.st
2023-02-10T10:24:01.637425Z
1 likes, 0 repeats
@cyclonebee @grey @InceptionState @Yeet okay thank you. like I said I will eventually play with it when I have time. I'm on a major time crunch this week to bring our operating costs down because the rest of our bills are due in a week but next week I am planning on doing absolutely nothing so I will learn this
(DIR) Post #ASXkjOnMr1MadYL4UK by eee@poa.st
2023-02-10T17:30:41.727675Z
0 likes, 0 repeats
@grey @InceptionState @graf @Yeet elf catgirl office lady gf?!