Stable Diffusion AI makes your bad drawings amazing - and you can download it for free
HTML-код
- Опубликовано: 11 сен 2024
- All examples done with the v1.4 weights, downloaded from HuggingFace. Stable Diffusion download links:
main repo: github.com/Com...
optimized for Nvidia cards with only 4GB of VRAM: github.com/bas...
Fork with M1 Mac support: github.com/lst...
Intel CPU support: github.com/bes...
You'll need to separately download the weights (about 4GB): huggingface.co...
If you need a GUI, use gradio (see README in repo), not the crappy GUI I made.
Latent Diffusion paper: arxiv.org/abs/...
What an amazing work, props to the authors. Thanks a lot for sharing, gonna give it a try.
absolute genius! simple explanations that i even understood
Wait... you're telling me that all those memes laughing about shows/movies going "enhance image!" is no longer a meme because it's possible now?
As long as you don’t mind that all the detail is made up!
where have u got with the valo ai
Just found your channel and love your content! You got my sub!
I'm speechless on how this is even possible
nice vid
Is it poss to run stable diffusion from the web browser instead of downloading it
Can I invest in you / your RUclips channel? Awesome content ...
I find the full release somewhat lacking. The lite version seemed to know much more, you could give it a prompt like "The Family Guy" and it knew it but this version doesn't anymore.
Yeah, I think the selection of training data really limited it, for better or for worse.
can someone confirm if this model has any malware (virus) embedded in it or not?
Can i run this on my gtx950 with only 2 gb of vram? Yeah, it's not so much but I am ok to wait after night to wait for the result if the problem is only in time of processing. Or there are other problems which wont let me even run this?
I think the answer is yes, if your card supports versions CUDA higher than 3.7 and the current pitorch. Why don't you try? You can also run sd on cpu only. But both methods will be quite slow. I tried to use my gtx 650, but the it's CUDA version is old and the sd does not start. Generating one image 512x512 on my PC with xeon x3470 and 8gb ram takes about 12 minutes. With modern processors, this time is 1-5 minutes. There is a video on youtube. Search RUclips for usernames
1littlecoder, darkhemicdigitalart, this will be helpful.
@@viacheslavmakarenko8411 wow, nice, thanks, i'll try it. I have Xeon E3-1275 and it should work +- as yours, and 12 minutes for me is ok, i worked a little with blender and you know, to get an image took hours :D
Try this for CPU inference: github.com/bes-dev/stable_diffusion.openvino
@@riveducha
Thanks! I love being able to throw a half-done render into SD for when I need something super quick n dirty, so this is a big help!
That said, holy shit those Zelda pictures look fucking *awful* XD
Haha cool :)
I don't understand the drawing part. You basically draw on an app (paint ?) then upload the image in stable and remake it through the algorythm ?
Right, and the initial image can come from anywhere, like photos or video game screenshots.
@@riveducha witch feature? Img2img? On stable diffusion
i dont have an nvidia card... cries
There's probably gonna be an amd update in a few months. You could buy a used 1660 for almost nothing.