Add in-process FastVideo executor (no worker spawn) for ZeroGPU

Speed: text-embed cache + lighter 384x640 profile + PyAV (VAE compile opt-in)

AOTI ZeroGPU-clean: pull prebaked bf16 .pt2 from HF bucket (aokit), no runtime compile

Update NVFP4 notes: flashinfer AOT-dir + HF bucket, speed/memory table

NVFP4: use flashinfer AOT dir + prebaked .so from HF bucket (no JIT, relocatable)

NVFP4 (FP4 DiT) opt-in path, ZeroGPU-clean via prebaked flashinfer kernels

Wire AOTInductor compilation of the LTX-2 DiT (aokit) into the live pipeline

Fix edit steering: single stream_start + extendable target

Add DREAMVERSE_MOCK mode for GPU-less control-plane testing

Continuous auto-rollout + rewrite steering in /ws (DreamVerse streaming mode)

Serve the REAL DreamVerse frontend + /ws streaming backend on gradio.Server

gradio.Server demo: real LTX-2 t2v backend + DreamVerse UI (ZeroGPU-shaped)

Add DreamVerse-styled UI (index.html) for gradio.Server demo

Replace stand-in with REAL LTX-2 DiT AOT artifacts (cu128 + cu130)

Add DreamVerse->aokit AOT compilation adapter + compiled pt2

Rename app.ph to app.py

Create app.ph

initial commit