Upload folder using huggingface_hub

Upload folder using huggingface_hub

Quantize Gemma before loading the rest: transient bf16 peak OOM-killed startup on tightly packed hosts

Tighten ZeroGPU duration: scale with pixels x frames (measured 0.33s/M)

Quantize Gemma (int4 QAT / int8 fallback) to fit default ZeroGPU slice; drop xlarge

Request xlarge ZeroGPU slice (53G packed ledger OOMs the default slice)

Dynamic ZeroGPU duration (cold-worker warmup blows fixed 75s budget)

Use Lightricks' ungated Gemma mirror (no license gate, no token)

Upgrade to LTX 2.3 22B distilled (Hub-downloaded checkpoints) + recipe

Apply Upsampler space recipe + standardized metadata

use xlarge ZeroGPU (full Blackwell) for VAE-decode headroom

add note about the update to v0.9.8 (#24)

update to 0.9.8 (#23)

mcp-improve (#22)

Update app.py

disable deeplink temporarily

update-inference (#4)

Update app.py

oops

get rid of gr.State

Update app.py

Randomize seeds sby default

Fix `improve_texture` and hide video-to-video

Update app.py

Update app.py

Add DeepLinkButton (#2)

MCP support? (#1)

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py

Update app.py