Top: VTM 1.5 | Bottom: 1.0
VTM-1.5 prototype training is done Testing shows better hair and clothing but face quality dropped significantly. One possible cause is bad data we noticed but didn't remove. We'll fix it in next prototype though it may not be the only issue #AIart️️
VTM-Elf v1.5 is coming soon! 3× more data (45K → 135K), plus accessory labeling + hair masking for better tracking and smoother movement. We’re finishing the data and Labeling now. #VTMElf#VTuber#AI#Diffusion
VTM Elf, is !OUT a side project that turns a reference image into a VTuber avatar. 30M param model, 12-13 FPS on my RTX 5060 Ti. Still blurry, but you can tell it's your character. Next: 3x the data
https://t.co/0cUmvcAEaL #VTuber#AIArt#StableDiffusion#IndieDev
🚀 VTM-Elf update: major progress! Skeleton mapping is working, movement still needs refinement, and speed has improved. Next step: advanced diffusion distillation with a larger model. #Diffusion#AIArt#VTuber#ImageGeneration#AI