Preserve MMAudio stereo; stereo-wins channel normalization

Fix (1,T) shape crash in _cf_join; bump HunyuanFoley secs_per_step to 0.5

Free GPU memory between HunyuanFoley segments to prevent OOM

Revert to single multi-seg GPU call; bump hunyuan load_overhead to 90s

Fix HunyuanFoley initial gen: one GPU call per segment

Fix HunyuanFoley: pre-download SigLIP2, use local_files_only=True

Fix MMAudio: load BigVGAN from local snapshot dir, not HF network

Fix HunyuanFoley: save text_feats to disk inside GPU worker

Fix HunyuanFoley xregen: register model lib sys.path in main process

Fix HunyuanFoley xregen AttributeError: return torch.device from _get_device_and_dtype

Add try/except traceback logging to _hunyuan_gpu_infer to surface AttributeError

Fix TARO GPU abort: increase load_overhead 15→120s to cover model loading time

Add granular step-by-step logging in _taro_gpu_infer to find exact GPU abort point

Fix pending regen waveform segment boundary misalignment

Fix MMAudio TypeError: cast num_steps to int in _mmaudio_gpu_infer

Fix TARO GPU worker crash: use local_files_only=True for audioldm2 VAE/vocoder

Add xregen error logging: try/except in _xregen_dispatch, logging in xregen_mmaudio._run and _mmaudio_gpu_infer to surface GPU worker crashes

Add verbose debug logging throughout TARO generation pipeline

Fix xregen ZeroGPU TypeError: duration fns must accept all positional args

Fix ZeroGPU tmp file isolation: extract clips inside GPU worker

Fix ZeroGPU kwarg stripping: make silent_video/segments_json positional

Fix 'Too many arguments': use **_kwargs in duration fns for extra params

Fix regen/xregen errors when silent_video tmp file is missing

Fix waveform crossfade hatch + click hit-detection to use contact edges

Fix ZeroGPU isolation for MMAudio/HunyuanFoley regen + waveform contact-edge fix

Fix NameError: restore import html as _html used in _build_waveform_html

Refactor: reduce technical debt across app.py

xregen: use target model's optimal window centered on segment midpoint

Remove pre-truncation from all inference paths — stitch_wavs owns trimming

Fix segment bleed: equal-spaced windows + contact-edge trimming

Fix regen GPU fns: move seg clip extraction inside GPU scope

fix: remove ctx_key from all function signatures — use fn-name-keyed global dict

fix: use UUID-keyed global dict + ctx_key param for ZeroGPU context passing

fix: pass CPU context as ctx_json argument to @spaces.GPU functions

fix: replace threading.local with caller-thread-id dict for ZeroGPU context passing

refactor: replace _cpu_ctx with thread-local storage, deduplicate xregen wrappers, parallel downloads, quiet=True

fix: pre-download BigVGAN vocoder to avoid GPU budget drain on MMAudio cold start

fix: pre-download cvssp/audioldm2 at startup to avoid GPU budget drain

fix: use stream selectors instead of map list in mux_video_audio

debug: expose ffmpeg stderr in mux_video_audio

transcode_for_browser: write H.264 into same upload dir so Gradio file server allows it

transcode_for_browser: return new path in fresh tmp dir to beat Gradio probe race

transcode_for_browser: overwrite in-place to suppress Gradio fallback converter

Fix: map 0:v:0 explicitly in transcode and mux to skip data/timecode streams

Fix: restore transcode-on-upload for browser preview; mux already re-encodes so no conflict

Fix: re-encode video in mux_video_audio instead of stream-copy to handle any input codec

Fix: replace custom transcode with gr.Video format=mp4; remove upload handler

Fix: handle no-audio video in transcode_for_browser; fallback on error

Fix: transcode uploaded video to H.264/AAC on upload for browser preview

Fix: restore regen duration floor to 60s (was regressed to 20s)