|
Do you think someone will make an easy way to install this locally? Or will you always need to know python and stuff to do this on your own machine?
|
# ? Aug 22, 2022 20:36 |
|
|
# ? May 29, 2024 02:36 |
|
A Strange Aeon posted:Do you think someone will make an easy way to install this locally? Or will you always need to know python and stuff to do this on your own machine? It will always need python because that's the core, someone will likely make an easy to use installer package that will set it up and a front end UI on top of all of this almost certainly. The devs here have said a few times they aren't looking to make an end product but something that will be part of end products but something that is also open source. I'm going to see how hard this is to get running on an RTX 3070 mobile 8GB. It's in a Legion and has the full VRAM, might get thermal limited, I'll have to see how much of a space heater this makes the laptop. What I kind of want to do is make a script that will take something simple like A penguin eating a coconut. Then have it make a few hundred iterations with variations of , 4k, matte painting, digital painting and other common tags. Each with various -C values as well Let the job run for the 2-3 hours and just look at the thumbnails of the 500 or so images and iterate on the most interesting. - something you could never do with a discord or website.
|
# ? Aug 22, 2022 20:42 |
|
Maybe stupid question, but why are they releasing a free open source version? I assumed their business model would rely on safeguarding their intellectual property. Is the open source version inferior?
|
# ? Aug 22, 2022 20:48 |
|
Rinkles posted:Maybe stupid question, but why are they releasing a free open source version? I assumed their business model would rely on safeguarding their intellectual property. Is the open source version inferior? Their stated goal has always been to be open source and deliver AI to the common person (which is also why it runs on gaming hardware). They don't want companies in control of it. They will be used by major projects such as MidJourney, which has already shown interest in integrating stable diffusion into it. I know it's weird, it's like 90s internet it's released because it's cool. It's hard to remember and wrap your head around again. That or they are going to mine so many bitcoins on all these GPUs.
|
# ? Aug 22, 2022 20:53 |
|
Yeah it was always planned to be open source from the jump. That being said they partnered up with LAION for funding and I'm sure they did a poo poo load of data mining on all the prompts people did. They can probably also make money incorporating non-model improvements into their Dream Studio frontend. Things like better language parsers, chaining multiple models together, nice inpainting and collage tools, etc. All that stuff is things you could do if you have strong comp. sci. skills and know how to work with all the python ML frameworks (there are a million) but most people just want to drive a car, not build it first.
|
# ? Aug 22, 2022 21:21 |
|
They're already charging you for the website version, so they'll probably just keep working to expand and monetize that further. Running it on your RTX card is a pretty niche thing for some turbonerds I'd imagine. pixaal posted:It will always need python because that's the core, someone will likely make an easy to use installer package that will set it up and a front end UI on top of all of this almost certainly. The devs here have said a few times they aren't looking to make an end product but something that will be part of end products but something that is also open source. The "optimized" model I tried so far basically unloaded and quit after every batch but I'm sure you could just set up all the permutations you want in the python script directly and keep everything in-memory for every prompt you try.
|
# ? Aug 22, 2022 21:29 |
|
also apparently OpenAI (Dalle2) is slashing their prices already. just some big coping lol
|
# ? Aug 22, 2022 22:11 |
|
Popoto posted:also apparently OpenAI (Dalle2) is slashing their prices already. just some big coping lol Yeah got an email from them about a half hour before the public release of Stable Diffusion.
|
# ? Aug 22, 2022 22:37 |
|
for anyone who wants to play with Stable Diffusion without worrying about credits: it's now up on hugging face https://huggingface.co/spaces/stabilityai/stable-diffusion
|
# ? Aug 22, 2022 22:56 |
|
Objective Action posted:Yeah it was always planned to be open source from the jump. That being said they partnered up with LAION for funding and I'm sure they did a poo poo load of data mining on all the prompts people did. Someone will build the pieces slowly as open source projects and someone else will bundle it into an easy script then someone else will bundle that into an installer and bolt a Bitcoin miner to it and mine while it's idle for them not you.
|
# ? Aug 22, 2022 22:56 |
|
Is the stable diffusion Image2image any good? It didn't seem to work well when I was trying it out but I'm wondering if its only useful for certain use cases.
|
# ? Aug 22, 2022 23:08 |
|
TIP posted:for anyone who wants to play with Stable Diffusion without worrying about credits: it's now up on hugging face Thanks. Seems you have to change the seed if you want a full reroll. These used the same seed surreal polish movie poster of dystopian ant colony surreal 70s polish movie poster of dystopian ant colony
|
# ? Aug 22, 2022 23:14 |
|
It wasn't part of the beta test at all, imagine it's pretty early development. DISCO Diffusion's Image to Image is also pretty meh (outside of portrait mode which is special settings for head shots, that nails it) so maybe try people and then add a job title like Portrait of the Queen of England 2082 with a picture of whoever you want and see what you get. I'm not sure how related the behind the scenes stuff is with DISCO and SD I could be sending you in the wrong direction.
|
# ? Aug 22, 2022 23:19 |
|
All the Bitcoin miners are repurposing their GPU farms to enable mass AI porn generation, humankind wins the game through cultural and technological victory
|
# ? Aug 22, 2022 23:20 |
|
wait times seem completely random for me. I've had image generation take 6 minutes, whilst another prompt I ran simultaneously took 15 seconds.Rinkles posted:Thanks. Seems you have to change the seed if you want a full reroll. this is actually useful for iterating on an idea
|
# ? Aug 22, 2022 23:27 |
|
There’s a fork of txt2img that spins up a local web server, automatically randomizes the seeds, and does not unload the model after generations: https://github.com/harubaru/waifu-diffusion/blob/main/scripts/txt2img_gradio.py Just download the script and place it in the script directory. Run ‘conda activate ldm’ and then ‘pip install gradio’ to install the gradio dependency the script needs. Then run the script. Works fine with the newest 1.4 model.
|
# ? Aug 22, 2022 23:29 |
|
Popoto posted:also apparently OpenAI (Dalle2) is slashing their prices already. just some big coping lol I think that's just the prices for textgen
|
# ? Aug 22, 2022 23:39 |
|
bobfather posted:There’s a fork of txt2img that spins up a local web server, automatically randomizes the seeds, and does not unload the model after generations: Yep this is essentially the premeio way to use this right now, slick and easy. Thanks for this. 5~ seconds to generate incredible art, in the hands of anyone.
|
# ? Aug 22, 2022 23:51 |
|
I'm mostly excited for when we can get good tools to do selective (re)inpainting. The have so many pictures that are cropping off faces or have hosed up hands that I could fix with that.
|
# ? Aug 23, 2022 00:48 |
|
I've tried getting SD to run locally, but I keep getting this error: RuntimeError: CUDA driver initialization failed, you might not have a CUDA gpu. Am I just turbo screwed? I've googled for hours about this and turned up with nothing (at least that I can comprehend). I've got an NVIDIA GeForce GTX 1060 with 6 GB VRAM.
|
# ? Aug 23, 2022 00:50 |
|
Yoshi Jjang posted:I've tried getting SD to run locally, but I keep getting this error: I ran it fine on a 1070. Maybe something hosed up with your drivers, try (re)installing the latest version
|
# ? Aug 23, 2022 00:54 |
|
from this one https://huggingface.co/spaces/stabilityai/stable-diffusion A photograph of a mermaid eating a lobster
|
# ? Aug 23, 2022 01:23 |
|
Rinkles posted:wait times seem completely random for me. I've had image generation take 6 minutes, whilst another prompt I ran simultaneously took 15 seconds. It gives you a time estimate that also shows a queue, so when it takes longer it's because multiple people are waiting ahead of you. Thankfully the estimate seems to be pretty accurate.
|
# ? Aug 23, 2022 01:34 |
|
bobfather posted:There’s a fork of txt2img that spins up a local web server, automatically randomizes the seeds, and does not unload the model after generations: For some reason, I cannot get this to work. I've followed all the instructions several times, but I keep getting different errors. Luckily, I can still get the regular way to work, so I can post madness like this. fist fighting a mountain, claymation styke, sinister tone Even with the misspelling, this looks amazing, especially when you realize AI made it. I also tried it with the misspelling fixed to see if anything would change, which it really didn't.
|
# ? Aug 23, 2022 01:40 |
|
8 images take about 2 minutes on my 3070 8GB Mobile. This is fast, you can modify kdiff.py to increase the max -C and -s values in that web UI. I tried to get fancy and changed the sliders to numbers for input boxes, this was a mistake that also changes it from an int to a float (you'll get an error back in anaconda saying as such). So only touch the max value and maybe steps if you want that too.
|
# ? Aug 23, 2022 01:58 |
|
I can't seem to get the newly released Stable Diffusion running on an AMD card. Even with the AMD/CPU installs of Torch I get things complaining about not having Nvidia GPUs/cuda cores.
|
# ? Aug 23, 2022 02:04 |
|
Got it running locally on a 2070S, both the command-line and Web UI versions. I can run the standard version, but I am limited to 512x512. These were generated with the optimized SD. "woman standing in front of a futuristic city, art by Roger Dean and Thomas Kinkade" edit: yes, i live in the booty-lookin arcology to the right. can't miss it. VectorSigma fucked around with this message at 02:19 on Aug 23, 2022 |
# ? Aug 23, 2022 02:16 |
|
"Calvin and hobbes real life, pastoral suburbs, realistic" is bizarre, especially whatever that is on the bottom left. "State of Liberty vs the Stay Puft Marshmallow Man, new york city skyline, realistic" didn't give me much, but it gave me this little guy. "Robocop vs Godzilla" plus some other modifiers gave me this weirdness Some of these designs are kinda cool and I especially love ROPM TOPUBLE
|
# ? Aug 23, 2022 02:20 |
|
I found a fun use for the image prompt feed it one you really like then provide basically the same prompt you can set the noise level to like 0.60 can get vaguely the same or something like 0.30 to get small variations. So if you really want to work part of the image you can use image to image and tell it to make 200 iterations quickly. Well 200 at 4 images a minute is an hour but hey still not bad! I'm going to forget to clean this output folder and wonder why I'm out of storage in a week aren't I?
|
# ? Aug 23, 2022 02:41 |
I really like the look it gives when using godzilla as a prompt. They are so noisy they really look like they are film stills scanned from a 80s magazine. all supposed to be Godzilla fighting Stay Puft Marshmallow Man lunar detritus fucked around with this message at 02:45 on Aug 23, 2022 |
|
# ? Aug 23, 2022 02:42 |
|
"a corgi wearing an astronaut suit, dragoncon, 35mm, f/2.8, flash fired, full length"
|
# ? Aug 23, 2022 02:43 |
|
I finally got that new script working (didn't realize it was a gui and that you just ran the script), but yeah, I'm digging this new system. "The gateway between dreams and nightmares, sinister magical vibe, 4k, realistic, vray lighting"
|
# ? Aug 23, 2022 03:31 |
|
LifeSunDeath fucked around with this message at 03:50 on Aug 23, 2022 |
# ? Aug 23, 2022 03:39 |
|
lunar detritus posted:I found this neat collab, that basically asks CLIP how it'd describe an image. Then, you can use that prompt to generate your own. Pretty decent! -->
|
# ? Aug 23, 2022 03:55 |
|
this guide worked for me for getting this running locally on Windows my av!
|
# ? Aug 23, 2022 04:41 |
|
Mozi posted:this guide worked for me for getting this running locally on Windows That's really sweet! I might look into this given the 1080ti I have sounds like it could be beefy enough, even if it's not as quick as a newer GPU.
|
# ? Aug 23, 2022 04:51 |
|
Mozi posted:this guide worked for me for getting this running locally on Windows thanks for the link. masterpiece_vermeer_painting_of_a_bear_sipping_coffee_smiling_at_the_camera_in_a_victorian_room ~21s per picture on a 3060ti, but it's only using ~4GB of VRAM (the 8GB command won't run because I only have 8GB total).
|
# ? Aug 23, 2022 06:01 |
|
So I found a real good AI for fixing hosed up faces made by other AIs. Here is the result of it fixing a couple of Midjourney portraits: I also found some really lovely old art I made, and it also worked pretty well on that:
|
# ? Aug 23, 2022 06:02 |
|
some of the 2077 AMG models
|
# ? Aug 23, 2022 06:31 |
|
|
# ? May 29, 2024 02:36 |
|
Good job, Stable Diffusion
|
# ? Aug 23, 2022 06:33 |