I’ve tried several AI projects from other bros and found they aren’t really up to par.I looked through a number of videos and noticed YOLO still has some issues with detection.Recently, I saw an open-source model called SAM3, and its image segmentation technology looks really solid.I’m wondering if we could use it instead of YOLO to handle tracking and then generate scripts.What do you guys think?
go ask fungen guys they are into that stuff
Recently, I posted a question on their GitHub and haven’t received a reply yet—they probably missed it. I just tested SAM3 and SAM2 on a cloud platform. The performance of SAM3 impressed me, but I’m not sure if it can identify specific objects accurately. Also, it consumes a lot of GPU memory, so video processing might require writing separate logic. On the other hand, SAM2 requires manual annotation, but I think its performance is still decent. By combining human corrections on top of SAM3’s outputs, I believe the results could be highly effective.q
There is also
The folks over at Fungen haven’t gotten back to me. They’re probably busy with other things.
but,I’ve now built a simple project that generates funscripts from SAM3. The scripts generated so far have keyframes that match the movements in the video, but the amplitude is quite low. There might be minor misalignments at some points, but it’s not a major issue and can be fixed manually. This is way faster than writing scripts from scratch—you only need to adjust the amplitude and remove the abnormal keyframes.
However, I have no plans to release this project. It’s just a simplified version that cannot generate perfect scripts directly. Its usage and configuration are somewhat complicated, and I’m not very fond of writing documentation.
eeeerm go to their Discord?
Sorry to say but the FunGen project is dead at this time.
Have you tried using MTFG (Moton Tracking for OFS)?
Is it better, worse or similar, If you have?
I’ve tried this one too, but the result wasn’t very satisfactory. In my opinion, SAM3 works better than this one—at the very least, its rhythm of ups and downs is consistent. 