Add ftfy (Wan V2V/VACE/I2V/Animate prompt_clean needs it)

Wire V2V + FLF2V + TI2V-5B (batch 1) — prepare→generate, mirror-aware load

Enable Xet fast-transfer; revert default to Wan 2.1

Prefetch retry on CDN drops

MoE .to(cuda) on xlarge (bf16 fits 96GB)

No download inside metered GPU call: startup prefetch + non-GPU prepare step

I2V large/xlarge routing

Device-aware bf16 loading + T2V large/xlarge slice routing

Load weights from HF CDN (fast) instead of the slow volume mount

Stabilize: gate startup load off (restore RUNNING)

Module-level startup load + default Wan 2.1 14B + torchao 0.17 pin

Disable background warm (restart loop); lazy-load with adequate budget

Post-launch background warm of Wan 2.1 T2V-14B (COW for workers)

Cold-load-aware ZeroGPU duration budgets

Stabilize startup: lazy model load + wan2.1-t2v-14b stitch bundle

README: proper OSS landing page + Apache 2.0 LICENSE

Stitched-dir loader — zero downloads via mount + bundled JSONs

Disable track_tqdm — UI scroll growth fix attempt

Preload T2V handle; split load from CUDA-attach

Pass local snapshot path + worker probe

Redirect HF cache to /tmp + visible snapshot_download

Trim preload to 1 model + LoRA (150GB cap)

Bypass volume mount + add base model preloads

DEBUG — dump mounted-volume contents at startup

Fix ZeroGPU 422 — static size literal + WIP dev banner

Wave K — mobile + tablet responsive (Linear theme preserved)

Phase 1 fix — drop MoE to large + CPU offload (PRO compat)

Wave I — port 7860 on HF Spaces (commit d92b33a)

Wave I — move theme/css to gr.Blocks() ctor (commit d7ab412)

Wave I — import spaces first to fix CUDA-already-initialized (commit 6e2d426)

Phase 1 — Wan Studio T2V/I2V on Wan 2.1 + 2.2 (commit fec10de)

Bump gradio to 5.49 (fix HfFolder import error)

Initial probe: detect ZeroGPU hardware (H200 vs Blackwell)

initial commit