Switches to the F1 model (different DiT path and logic).
If checked, 'Input Image (Video Start)' is hidden. Each batch item uses a random image from the folder.
How the end frame affects generation (if provided)
Resolution Options (Choose One)
Define specific prompts and starting images for different sections of the video. For the index you can input a range or a single index. A 5 second default video has 4 sections. The first section is 0 and the last is 3
--- Control Slot 1 ---
--- Control Slot 2 ---
--- Control Slot 3 ---
--- Control Slot 4 ---
Enable FP8 precision for the main Transformer model.
Enable FP8 for the Llama text encoder.
Requires FP8 DiT. Use scaled math (potential quality improvement).
Decode all frames at once instead of section by section.
Enable TeaCache for faster generation (shits hands).
Uses og model supports end frame. Default is F1 model.
If checked, the VAE latent of the guidance image will be used as the initial conditioning for the first generated segment. Turn down context frames when using this
Core Generation Parameters
If checked, only the newly generated extension part of the video will be saved.
LoRA Configuration
Select model type. *-FC options enable Fun-Control features
For mixing fp16/bf16 and fp8 weights
Select model size: t2v-1.3B is faster, t2v-14B has higher quality
For mixing fp16/bf16 and fp8 weights
Model size: t2v-1.3B is faster, t2v-14B has higher quality
For mixing fp16/bf16 and fp8 weights