Replace FlashSR with sinc resampling for ZeroGPU compatibility

Fix HunyuanFoley CLAP tokenizer overflow for long prompts

Fix FlashSR CUDA init: bypass FASR class, load SynthesizerTrn on CPU

Optimize ZeroGPU: move FlashSR to CPU, fix regen 16kHz bug

Move FlashSR inside GPU window, fix xregen routing, refactor helpers

fix: load FlashSR on CPU only to avoid ZeroGPU CUDA init violation

remove: HF token settings section and JS fallback

feat: add FlashSR post-processing to upsample TARO 16kHz → 48kHz

fix: remove CF label from crossfade hatch indicator

fix: move segment color boundary to crossfade midpoint

fix: centre crossfade hatch on color boundary, not after it

fix: correct crossfade indicator position on waveform

feat: add crossfade overlap indicators to waveform display

fix: cap crossfade slider max to 4s and clamp in _build_segments

refactor: extract shared helpers to reduce technical debt

Optimize GPU usage and fix resource leaks

Use Gradio's postMessage zerogpu-headers for regen quota attribution

Fix crossfade loudness bump: apply db_boost to overlap as a whole

Add HF token input for ZeroGPU Pro quota on regen

Fix ZeroGPU Pro quota: fetch fresh x-ip-token before each regen call

fix: relay x-ip-token from HF proxy to JS for ZeroGPU Pro quota attribution

fix: add gr.Request to xregen_* for proper ZeroGPU quota attribution

revert: back to queue/join — call endpoint gives 404 Session not found

fix: switch regen to /gradio_api/call/{api_name} for proper ZeroGPU quota attribution

fix: restore waveform HTML before writing error to statusbar

fix: align defaults with official repos

fix: add credentials:'include' to regen fetch so HF cookies are sent

feat: show quota error in status bar instead of toast

fix: revert to document.body for toast — window.top blocked cross-origin

fix: use window.top.document for toast and max z-index

feat: show quota/error toast and restore UI state on failed regen

debug: log SSE data shape and fix misleading success:false log

fix: replace render=False xregen outputs with real rendered slot components

fix: cross-model regen — fix stereo/mono mismatch and add pending spinner

feat: cross-model segment regen — regenerate any slot with TARO, MMAudio, or HunyuanFoley

fix: free GPU memory between samples to prevent VRAM fragmentation

fix: use render=False on regen buttons to suppress Gradio 5 SSR 'Too many arguments' error

fix: move MMAUDIO_WINDOW and HUNYUAN_MAX_DUR before MODEL_CONFIGS

fix: allow Gradio to serve /tmp files so waveform audio loads (403 fix)

refactor: consolidate duplicated code via MODEL_CONFIGS registry

perf: move CPU work outside @spaces.GPU to reduce ZeroGPU cost

perf: comprehensive optimization pass — caching, dedup, cleanup

fix: also hide video download button via CSS as belt-and-suspenders

fix: hide built-in Gradio video download button on output slots

fix: add video download link to waveform panel so it always reflects latest regen

fix: draw and hit-test segments by unique region to prevent overlap highlight bleed

fix: derive total_dur_s from ffprobe not CAVP frame count to prevent audio truncation

fix: load only AudioLDM2 VAE+vocoder subcomponents instead of full pipeline to prevent GPU OOM on long videos

Fix seed: default to -1 (random) instead of baked-in startup value

Randomize seed fields on every page load via JS