@zrottmann@victormustar It doesn’t matter what animation, the key is to have a valid video - 2 same moves each frame from video generator.
Reader can do mistakes when some limb is hidden from both perspectives, then it’s guessing.
Can be improved further
NVIDIA's MotionBricks makes every frame between a few poses, live, while the game runs. No animation clips. I tested it with our own game animations. Full video: https://t.co/GFSBc0GB90
#gamedev#AI
Opus wrote the rest in C++: the boss AI, real hitboxes, the hero's motion matching, Niagara fire.
The tool is free and open source:
https://t.co/8f1TaZEGHj
Full video: https://t.co/YZActvipbx
5. That reader goes through the AI video frame by frame, finds every joint in both views, and puts the same move on the boss's own skeleton.
Out comes a normal Blender animation → straight into Unreal.
@maxt3chno You can try my new tool, not sure if you saw my new video about it, it’s much better and much more token efficient :) it’s also on GitHub - video2rig
Opus 5.5 vs GPT-6.1 Sol, same $200 each: a cyberpunk shooter inside a cube where gravity turns at every edge. Opus built its city kit with 1 agent in under 2 hours. Sol ran 14 agents for 13 hours.