Use Noto Sans font for universal language support

Fix Whisper transcription: use audio_track instead of video clip

Add debug logging for subtitle path and FFmpeg command

Two-pass approach for subtitles and watermark

Fix subtitle visibility with watermark

Fix watermark transparency: use gr.File instead of gr.Image

Fix PNG watermark alpha channel handling

Fix watermark: add -loop 1 -t duration flags (from pycaps) to properly loop PNG over video

Move Whisper inside main GPU session (90s) - no separate GPU call, all in one reservation

Fix PNG alpha: use scale2ref filter (from pycaps) for proper transparency handling

Move Whisper to GPU session (30s reservation) - 20-30s for 10min audio vs 30-40min on CPU

Fix PNG transparency: add format=auto to overlay filter for proper alpha blending

Fix: Watermark overlays at 0,0 for full control - user positions in their PNG editor

Add Whisper subtitles (elegant animated) + PNG watermark support - all CPU work, free

Use separate input fields for 3 prompts instead of multi-line textbox

Remove unsupported info param from gr.Audio

Fix: Multiple prompts (one per line) for different clips, not variations of same prompt

Add multi-clip generation with audio looping + high res 1024x640: Generate 1-3 clips, loop to match audio duration (CPU work free)

Initial commit: Element-8 (vanilla LTX pre-distilled)