Spaces:

---

C4G-HKUST

/

AnyTalker

like 60

Running on Zero

Dear Users,

To ensure smooth GPU task processing, please adhere to these audio length guidelines:

Fast Mode (120s GPU budget, suitable for any type of users):

• Keep audio under 4 seconds. Audio inputs longer than 4 seconds will be automatically trimmed to 4 seconds.
• In multi-person mode, it's the total sum of lengths for everyone (for concat mode) or the maximum length (for pad mode).
• Fixed 8 denoising steps for quick generation.

Quality Mode (Dynamic GPU budget):

• GPU duration is dynamically calculated as: 60s (preprocessing) + video_seconds × denoising_steps × 3 seconds.
• For example, an 8-second audio with 25 denoising steps will take approximately 660 seconds (60 + 8 × 25 × 3 = 60 + 600 = 660s ≈ 11 minutes).
• You can adjust the number of denoising steps in advanced options to accommodate longer audio durations or reduce GPU time.

Longer Videos:

• For unlimited-length videos, please use our code on GitHub.