Revert torch pin (broke spaces/ZeroGPU + Wan2.2/LTX-Video); make diarization best-effort and drop pyannote.audio dep. Transcription-only stays GPU-accelerated.

Pin torch/torchvision/torchaudio to 2.2.2 trio for pyannote 3.4.0 compat (reverts the AudioMetaData shim, which didn't find the class)

Pin pyannote.audio==3.4.0 (newer versions redirect 3.1 pipeline to a different gated model)

Add audio transcription + speaker diarization tab (Whisper + pyannote)

Add Wan 2.2 pixel sprite animation endpoint

Upload requirements.txt with huggingface_hub

add txt2img tab + infer_txt2img endpoint (SD 1.5)

Add peft + torchvision (required by diffusers load_lora_weights)

Initial venture-studio deploy: pixel_cursor + studio + app.py

Duplicate from gradio-templates/text-to-image-gradio-template