Improve UI layout: fixed video height, align input/output with tabs

Revert VAE encoding to original sequential loop

Batch VAE encoding: encode all 5 G-buffer modalities in one call

Fix scroll button: use elem_id instead of gr.HTML

Add scroll-to-examples button, fix font consistency

Make Examples a clickable anchor link that scrolls to bottom

Add quick start hint pointing to examples

Add note about generation time and inference steps

Change default frames to 73

Adjust GPU duration to 180s

Change default inference steps to 20

Adjust GPU duration to 200s

Increase GPU duration to 240s

Optimize GPU quota: move CPU work before pipe.to(cuda), remove redundant pipe.to(cpu)

Super-squash branch 'main' using huggingface_hub