Uploaded March 2019 | Updated September 2026, 2 weeks ago

---

Danger Mouse (2015 Reboot) opening sequence upscaled to 8K (7680x4320) to showcase the capabilities of waifu2x.
[Open the description for FAQ and other useful info]

Warning: You will need a powerful computer to watch this video in 8K. I tried it on my computer and I was able to play it just fine, both on YouTube (in Opera) and locally (in MPC-HC)
If you don't see 1440p ("2K" HD), 2160p (4K) or 4320p (8K) then your device (non-desktop) does not support those resolutions, or you're using an unsupported browser (desktop)

If you think your computer can handle it in a video player app but not in a browser, or you simply want to avoid YouTube's compression by directly downloading the file I uploaded to YouTube, here you go:
[595 MB] mega.nz/#!K4ARSICS!LZM2yos9vyHeoCApeVCP5whjfe-r2PnNURCCJN6Iiyc



FAQ

Q1: Why?
A1: Why not?

Q2: How long did the upscaling take?
A2: 6 hours 45 minutes (GTX 970 4GB GDDR5, the original source is 1080p) - keep in mind that I was running an older version of waifu2x which did not require me to have CUDA/cuDNN set up. With the two installed and set up, the processing time is nearly cut in half.

Q3: What upscaling model did you use?
A3: RGB model and denoise level 3. Do note that UpRGB gives very similar results compared to RGB and is about twice as fast. CUnet (one of the two new upscaling models) is slightly better with the edges and takes about as long as RGB with the previous setup (older waifu2x on GPU without CUDA/cuDNN set up)
As a personal recommendation, I would use CUnet with TTA Mode for upscaling single images (for newer versions) or RGB with TTA Mode (for older versions, with/without CUDA/cuDNN set up) - as for video frames, either RGB or UpRGB.

Q4: How long did the render take?
A4: 14 minutes and 15 seconds (i7 6700K, VBR 2 Pass, 128Mbps target/192Mbps max)

Q5: How many frames were there?
A5: 972 frames (38.88 seconds at 25 FPS)

Q6: Where did you get the source footage?
A6: I'm afraid I can't answer that question, as we're talking about a full episode here.

Q7: Can you explain the whole process?
A7: Sure.
First, I used Premiere Pro to render the opening into a PNG sequence (for some reason, extracting the frames with ffmpeg, like I usually do, would produce a lower quality result, having jagged edges and altered coloring, probably because of the source format I had or something)
Then, I used Audacity to get the audio from the source; I trimmed the audio as I only needed the opening and made it 3dB louder (Making the audio too loud or normalizing it too much is a terrible idea, since the "loudness effect" will kick in and the audio will sound like it's blaring, even though it never peaks or is actually loud, which is why the source audio is so quiet in the first place) - for a better overview of this "loudness effect" I would recommend watching the first two minutes of 3kliksphilip's video about hearing - youtu.be/WRqVDb7sd0I
After that, I used waifu2x-caffe (github.com/lltcggie/waifu2x-caffe) to upscale/denoise each frame to 8K and let it run overnight.
When I woke up, I imported the frames as an image sequence into Premiere Pro (CC 2018 has support for exporting H.264 8K, which YouTube is happy with, unlike CC 2015, the one that I always use)
Finally, I made a sequence out of the image sequence and the audio and then exported the final result.

Q8: Dangerfold?
A8: I don't see how this is a question, but hell yes I ship it.



❯ Request videos
yoshifan.me/request

❯ Follow me on Twitter
twitter.com/YoshiFan13

❯ Other social media
yoshifan.me



❯ Rendered in Adobe Premiere Pro CC 2018



All rights belong to their respective owners. No copyright infringement intended.

This video may potentially trigger seizures for people with photosensitive epilepsy. Viewer discretion is advised.