Add first_image2/first_image3 per-clip frame conditioning

Use visible Textbox/Video components for prompt2/prompt3 and multi-clip outputs

Fix: add prompt2/prompt3 State inputs to generate_btn.click so API exposes them

Add multi-clip support (prompt2, prompt3 optional params)

Add music removal feature using Demucs (keeps vocals only)

Fix blur tensor format [F,H,W,C]->[F,C,H,W] and update all resolution defaults to 512x320

Default to low resolution 512x320

Add torchvision for gaussian blur

Add Gradio cache cleanup (delete files >2h old)

Fix: use torchvision.transforms.functional.gaussian_blur

Add Gaussian blur option (0=off, 36=heavy censoring)

Preload audio components (audio_encoder, audio_decoder, vocoder)

Fix preloading to use model_ledger attribute

Switch to DistilledPipeline with pre-distilled sulphur_distil_bf16 checkpoint

Use empty list instead of None for distilled_lora

Add required strength params (set to 0 since no LoRA)

Use fp8 without distill LoRA (incompatible with mxfp8mixed), increase steps to 30

Use fp8mixed checkpoint (23GB) instead of bf16 (46GB) to fit ZeroGPU

Preload stage 2 models including spatial upsampler

Fast parallel downloads with hf_transfer

Cache bust v2

Match Modal exactly: upsampler 1.0 + inference_mode workaround

Fix: use upsampler 1.1, default 896x512

Exact Modal config: sulphur_dev_bf16 + distill LoRA

Fix spatial upsampler to 1.0 to match Modal

Sulphur dev + distill LoRA with TI2VidTwoStagesHQPipeline

Sulphur + Lightricks 12B Gemma

Element-16 with Gemma 4B for smaller footprint

Element-16: fp8 optimized video generation

fix: fa3 broken on Blackwell+ (#10)

Update app.py

upgrade to [distilled 1.1]

Fix crash by pinning LTX-2 to a compatible commit. (#5)

Update app.py (#2)

add audio input support (#1)

Update app.py

Update app.py

Upload pinkknit.jpg

Update app.py

Update README.md

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update README.md

Update app.py

Update app.py

Update app.py

Update app.py