VaM 1.x Has anyone tried using AI (GPT, Claude, etc.) to create VaM animations?

Threads regarding the original VaM 1.x

leke

Member
Joined
May 6, 2024
Messages
33
Reactions
5
Hey everyone!

Recently, I've been experimenting with using AI tools like GPT (Astra 6 & Astra 6.2) and Claude (Opus5.5) to create animations in VaM. I'm mainly interested in making more detailed, polished, non-NSFW animations, with natural body movements, gestures, and expressions.

So far, I've tried two approaches:
  1. Having AI generate or modify VaM scene JSON files directly.
  2. Using AI to create animations in Blender, then importing them into VaM.
Unfortunately, neither approach has worked particularly well. I can get some basic animations working, but the results are usually pretty stiff and unnatural, nowhere near the level of quality I'm aiming for.

I'm wondering if anyone else has experimented with AI-assisted animation in VaM. Have you found any workflows that actually produce decent results? Maybe there's a better way to combine AI with Timeline, Blender, or other tools that I haven't thought of yet. I'd love to hear about your experiences or ideas.

I'm also curious whether AI is better suited for assisting with specific parts of the animation process, rather than trying to generate an entire animation from scratch. Thanks!
 
Last edited:
Kind of, I've tried with only local tools and in a round about way though. Basically, starting with a video generated with WAN 2.2 then put it through BlendCap to capture as an animation.

The BlendCap .bvh can be retargeted to a G2F/G2M rig using its own tools or through DAZ. I prefer retargeting to an MMD rig though as it seems to be easier to cleanup animation in MikiMikuMoving and then I can simply import into VaM with the mmd2timeline plugin.

I haven't tried doing anything with expressions yet as it would require making ARKit compatible shape keys.
 
Back
Top Bottom