Wow, this is really cool, looking forward to trying it out!
For anyone unfamiliar since the original post is pretty technical, this would let you take basically any VaM scene and render out _structured_ data in order for AI to interpret and generate an entirely new image based on your given prompt (for example.) I say that but I'm also aware that this community is no stranger to AI workflows.
One curious question: would it be possible to export the depth map and/or poses as videos in the future? That could be really useful for video models.
I actually think VaM and AI workflows complement each other super well. You can pose characters you have a LoRA for, or characters you inject via IP-Adapter, exactly how you want in VaM and then generate images or videos from it - I find that really cool.
Video export for depth / pose is a really good idea for video models like AnimateDiff / SVD. I can look into it for a later update if there is demand for it.
I also noticed that some clothing and assets are not correctly captured in the current Depth pass, because in VaM they come from different sources
(skin = opaque, clothing = cutout, hair = transparent - raw depthTexture only has opaque). That needs a fix anyway, so I will include the video export idea when I rework the depth capture.