Mouthless Yuki Nagato (AI GIF)BadBlinko on DeviantArt

Deviation Actions

BadBlinko's avatar

Mouthless Yuki Nagato (AI GIF)

By
Published:
1.9K Views

Description

It's a little misleading to just call it Ai, as it wasn't a simple one touch output. This took longer than most gifs I have done before it and it would have been easier to edit the gif as is. All for an 'interesting' art style. Just an experiment, not sure if I will do any more besides this.

(Edit: I might try again just to see if I could get any more interesting results or how to improve. I am unsure whether people actually like this or not.)

Original Found gif


Original + ai fps boost (not as good, not sure why it loops like that.)

Nagato Source-6x-RIFE-RIFE4.0-48fps by BadBlinko 

With AI fps boost of this edit (16fps). I went with original just because the smoothing between the frame with the ear out and neck down is too sudden to ever be fully smooth.

  Process 4-2x-RIFE-RIFE4.0-16fps by BadBlinko

------------------------------------------------------
(Edit: If you fiddle with denoising and prompt strength, you can get natural more consistent results (with lower being closer to original). The more you deviate, you may one soft inpainting and make sure it is blurry to get rid of bad deformation noise errors).

What I did (and it took me at little bit to figure how what to do)

Step 1: Open original gif in gimp and export as .ora to extract layers as independent pngs. 

They are in reverse order so you will have to reorder them by naming then something like Frame 1, Frame 2, etc... For a few steps I used different folders.

Step 2: Stable Diffusion Prompting

I spent a few hours to get the prompting right. For the eye closed frames you need to change it to eye close. Unfortunately in later readjusted frames I forgot to include "Yellow eyes" so some turned out more grey. I tried various "Nagato Yuki" lora to work with my model as well as found other lora to make it as consistent as possible (like consistent lighting Lora). This took a lot of trial and error. Unfortunately, the best outputs were not used as they were hard to replicate.

Step 3: Soft inpaint + Img to Img generation. 

I had to reload S.D to make it consistant for some reason. Basically I went frame by frame with img to img.

Step 4: Redoing frames for consistency

For consistency I had to regenerate particular frames over and over again. I focused as little errors as possible. The most are probably with the eyes unfortunately, but I trying to focus the colarbone (almost impossible by I tried my best), the hair shape, the glasses (often the eyes were treated as second eyes and had their own eyelashes, the uniform (was fixed when I used a good lora, but it would often change the scheme subtly: two stripes, different colors, etc...), art style, and best quality. I could probably get it to be better, but it would take a lot longer.

Step 5. Gimp editing + Mouthless edit

Open then as layers in gimp. I did minor color adjustments like chrome, noise reduction, etc.. to each layer. The background layer was distracting, so I removed the background on each layer.

Step 6 Export + Fps boost

I tried fps boost with Flowframes in itch.io. Its..... fine... just the frame transistion I mentioned above really kills it. I had to export as a .ora file again and reorder them in order to create a folder to use flowframes on. Otherwise, I would do my normal way of exporting gifs with highest quality (don't use gimp). Export from gimp as a loseless webp file and use ezgif.com webtogif converter

Most of the time was spent figuring things out, but it certainly was interesting and had an interesting output. 
-------------------------------------------------------

Let me know if this art style would be worth trying again. The key difference Is that I don't need to make the gif myself, which does take time (finding source and recording it or taking pictures frame by frame using an autoclicker.)

Image size
1360x768px 7.64 MB
Created using AI tools
© 2024 - 2025 BadBlinko
Comments0
Join the community to add your comment. Already a deviant? Log In