Switch between multiple GPUs, with queue functionality supported.
Set the maximum memory limit based on your own GPU memory to prevent overflow; for memories over 24GB, you can set it to 0, which means no memory limitation.
Switch to fp8 models, which I personally find to have better performance.
Mount multiple loras.
Lock seed functionality, and the generated videos will carry all relevant information for future reuse.
- Key point: If a good result is achieved with a specific seed, you can lock it and fine-tune the prompt or lora strength to improve the results further.
As long as there’s enough GPU memory, you can generate longer videos instead of being restricted by the desktop version's 1080p 5-second limit.
First and last frame functionality.
Intelligent multi-frame functionality.
Action migration (including pose/depth/line control), camera movement migration, and video redrawing.
TTS voice generation and voice cloning functionality.