- 3
- 10 036
rajb245
เข้าร่วมเมื่อ 15 ก.ค. 2006
Radiohead - Videotape (Mephistopheles Mix) [AI Generated Video]
My contribution to the conversation about this song, made using the tools I know best. That means a machine-learning model for stem separation and custom python scripts for prompting the image model and making animations.
มุมมอง: 624
วีดีโอ
Turtles All the Way Down
มุมมอง 1529 ปีที่แล้ว
Signals recorded using Ettus B210, visualized with Osmocom FFT gr-fosphor, and video captured with FFMPEG x11grab. Then the capture window was resized and dragged onto itself.
THX Deep Note Spectrogram / Waterfall Plot
มุมมอง 9K11 ปีที่แล้ว
Created with MATLAB. Audio sourced from another youtube user: th-cam.com/video/uYMpMcmpfkI/w-d-xo.html Spectrogram was created using 4096 point hanning-windowed-FFT, with 50% overlap between spectral bins. Everything is normalized to 0 dB maximum, and values below the nominal 16 bit dynamic range of ~96 dB are clipped. I do not make any claims of copyright of the reproduced sounds. This video i...
Amazing work man. Do you have a github repo or something for this? Would love to see how this was made.
thanks! the video was generated with a modified version of the deforum notebook: github.com/deforum-art/deforum-stable-diffusion/ the audio was split into stems using hybrid demucs: github.com/facebookresearch/demucs
Youd get more views if you AI in the title. Like Radiohead Videotape remix with video made with AI.
Interesting point, wonder if I can update the title now...thanks for listening!
Wow - just found this via your comment on the SM64 In Rainbows video - this is astonishing. Did you create that video as well as the remix? If so, unbelievable and this deserves many orders of magnitude more views! Only wish you incorporated the arpeggio sequence from the earliest versions of this song into your mix. But serious kudos, and thanks for a moving version of one of my favourite songs...
Yessir, I do AI for a living and used AI-based tools for both the audio track separation and the video. Having listened to remixes and live versions of this song over and over, there's more I'd like to have done with this track, but I just put it out into the world to get it off my mind. Otherwise I'd futz with it for months and forget about it, then no one would ever see it.
@@rajb245 I'd be fascinated to learn what tools and workflows you use to create imagery like this. I'm familiar with still AI imagery but this is a very compelling video and I'd love to see more works like this...
@@GrahaminKorea this *is* AI still imagery, but each frame's initial condition before the diffusion noise/denoise process is a slight zoom in of the previous frame. each chunk of frames has a prompt, then the prompt changes at various points in the song. the net result is that you continuously zoom into this fever dream. after all those frames are generated, its just standard tools that can encode a sequence of images into video and multiplex / encode in the audio source to make one video file for upload to youtube. check out here, deforum.github.io/ but let me caution you that this isn't really user friendly product
Found this by search! The internet is a vast place
Draws a penis
How deep is that note!
Is there some aliasing present in the frequency analysis? Looks like the plot is kind of reflected about 13kHz.
That behavior around 13 kHz is an interesting observation. As far as I know, that effect is in the original THX audio, because I used a 44.1 kSamp/s audio file and just calculated the spectrogram. Get the audio and run some spectral analysis yourself for comparison and post a reply video if you like!
it sure does look as aliasing
I wonder what this looks like on a logarithmic scale.
The Audience is Listening.... Some is scarred..... Some is Creeped.
This is great! Any way you could share the script? I am having trouble running the spectogram in a loop and develop over time.
I'll look for it, but maybe this helps: I think I did the spectrogram all in one go, i.e., I calculated the final picture you see first. Then, I reveal it line-by-line in a loop synched up with the sounds.
Raj Bhattacharjea Ah, good to know, thanks. I am looking to display a continuous audio signal input in a waterfall format.