Models Used

• model = wan2.1_vace_14B_fp16.safetensors

  (WAN 2.1 Video Diffusion – core generation model)

• lora = Wan21_CausVid_14B_T2V_lora_rank32.safetensors

  (LoRA – improves motion / temporal consistency)

• clip = umt5_xxl_fp8_e4m3fn_scaled.safetensors

  (Text encoder – prompt understanding)

• vae = wan_2.1_vae.safetensors

  (VAE – latent ↔ image conversion)

Workflow Summary

• Upload a video + set target aspect ratio / resolution

• WAN generates in ~5 second chunks (81 frames @ 16 FPS)

• Total loops:

• Supports any length video

• Internally loops ~5s segments

• Outputs final video at correct length + aspect ratio