Sorry about the bugs, I've just released an update. The site's music should no longer shatter your eardrums until after you touch the unmute button.
The videos are generated from random https://lexica.art prompts, with linear interpolation between two random seeds for each video, held at the same prompt, looped with ffmpeg filter_complex reverse/concat. Music from various creative commons / free sources.
- modified img2img to accept 2 prompts and 2 seeds (and a slerpradius and output video path)
- uses imutil.Video to create a video
- stochastic encode with noise, then decode, then add a frame to the video, for all n_iter
one thing i dont understand is why OP keeps calling `.half()` on their models. googled it and it seems to be a newish pytorch feature for "half precision" but couldnt get a clear answer on why you would want that. anyone care to share?
Comments
Sorry about the bugs, I've just released an update. The site's music should no longer shatter your eardrums until after you touch the unmute button.
The videos are generated from random https://lexica.art prompts, with linear interpolation between two random seeds for each video, held at the same prompt, looped with ffmpeg filter_complex reverse/concat. Music from various creative commons / free sources.
Source code at https://github.com/lwneal/duckrabbit/
Hosted on a single $7 node at https://www.digitalocean.com
i find it interesting that you only took 9 commits to do it as well (+244 -62 LOC) https://github.com/CompVis/stable-diffusion/compare/main...l...
breaking it out for others curious:
- modified img2img to accept 2 prompts and 2 seeds (and a slerpradius and output video path)
- uses imutil.Video to create a video
- stochastic encode with noise, then decode, then add a frame to the video, for all n_iter
one thing i dont understand is why OP keeps calling `.half()` on their models. googled it and it seems to be a newish pytorch feature for "half precision" but couldnt get a clear answer on why you would want that. anyone care to share?
I believe using .half() lets you run Stable Diffusion on graphics cards with less vram
Half precision models reduce memory requirements without really impacting the final quality much.
Wow this looks nice. Lots of focus on details here in the thread but don't let that distract you.
From the code [1] I take it that you're using polar interpolation on two random points around the original noise right?
This is very cool stuff, thanks for sharing!
[1]: https://github.com/lwneal/duckrabbit/blob/cb375ec7c6067bf805...
Great work!
Could you explain more exactly which Droplet/configuration you are using on DigitalOcean?