v1.0.0 — first stable release

AMD x lablab Developer Hackathon 2026 submission.

One English sentence becomes a 30-second cinematic reel, end-to-end on a single AMD Instinct MI300X. Every model in the pipeline is Apache 2.0 or MIT.

Pipeline (8 stages, sequential on one GPU):
- Director Agent  : Qwen3.5-35B-A3B (vLLM + AITER MoE)
- Character masters: FLUX.2 klein 4B (reference editing, no LoRA)
- Per-shot keyframes: FLUX.2 klein with reference image
- Animation        : Wan2.2-I2V-A14B (FBCache 2x + torch.compile 1.2x)
- Vision Critic    : Qwen3.5-35B reloaded, 10-label failure taxonomy
- Music            : ACE-Step v1 3.5B
- Narration        : Kokoro-82M (9 languages)
- Mix              : ffmpeg

Performance on MI300X: 25.9 min -> 10.4 min per 720p clip (2.5x cumulative).