Home
Creators
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

AI Animation - 2 experiments! (Patreon)

Published:
2025-07-11 20:16:33
Edited:
2025-07-11 20:26:08
Imported:
2026-06
Tags:

Content

Hello there!

I was browsing the official OtaKing discord™®© yesterday and noticed some AI conversation, so I recorded a video to try to see- in real time, no less - is AI capable to help with inbetweening or is my career safe? Here's the discord link: https://discordapp.com/channels/495039441753341952/495039441753341955/1392683916074160270

Anyway, take a look at the video and see what you think. Just a little Discord/Patreon exclusive brainstorm session for you all.

-Paul

Files

AI frame interpolation!

Comments

Greeky

EGG-SPE-RI-MENT!

JennaJ

This is a very interesting experiment, thank you for sharing! The way AI has been introduced at my day job is "We (bosses) went and bought this thing (without asking you) that was promised (by the sellers) to cut your work time (not replace you!) so you can get more done faster and concentrate more on the tasks AI can't do, and now we need your help to figure out how it can actually do it". In your cases it'd be immediately abandoned for the second case, but in first? It would've taken you 100min to draw the 10 frames, and took 20 min to generate + Xmin for you to check every frame and redraw the wonky parts, so there's possible time save depending on the X. However, whatever X is, that timesave seems unlikely to cover the cost of the AI tool unless it's given to multiple people in a big company producing a lot of animation...

Ted Schundler

Is that really AI? It doesn't really look better than optical flow frame interpolation. (maybe worse even) Optical flow can have difficulty with large jumps. I imagine color would help that program a lot, so it could figure out the motion better. You'll still get the interpolated colors in some places unless it's been designed specifically not to. Ah, it's based on a 6-year old paper - https://github.com/baowenbo/DAIN And it sounds like it's optical flow + attempting to be smart about occlusions, and that's the ML side. This is from when GANs were still the cutting-edge of image generation. (aka https://www.thiswaifudoesnotexist.net/ ) A lot has advanced since then. Though I imagine linework-only training data might be lacking, so while better could exist, it might not. (I'm not a fan of the AI tools generally, just random comments as someone that has been involved in computer graphics for a very long time)