While Runway and Pika still queue in the cloud, Alibaba Tongyi Lab has crammed an entire Hollywood studio into your RTX 3060. On July 26, Wan AI 2.2 went fully open-source, ushering in the era of “four-minute movies for everyone.”
Performance: the open-source ceiling
Across eight benchmarks—VBench, EvalCrafter and more—the 14 B build crushes closed-source rivals. The 1.3 B lightweight model needs only 8.2 GB VRAM and delivers a 5-second 480 p clip in four minutes on an RTX 4090; even a laptop 3060 can keep up.Features: the hexagon warrior
• Text-to-Video — type “whales under a starry sky,” get the scene.
• Image-to-Video — upload a product shot on white, receive a 360° spin ad in ten seconds.
• First-Last-Frame — give two keyframes, AI completes the entire camera move.
• In-paint & restyle — repaint, restyle, swap characters on the fly.
• Bilingual text FX — world-first animated Chinese & English subtitles.
• Auto soundtrack — beat-synced audio generated in sync with visuals.Use cases: from TikTok to Hollywood previs
– TikTok creators: 15-second viral scripts visualized instantly.
– E-commerce: one SKU photo → 30-second product story.
– Animation studios: text scripts become animatics in one click.
– Education: concept diagrams morph into experiment demos.
– Indie directors: preview entire long takes with two keyframes at minimal cost.Open-source freedom
Code, weights and the VAE codec are MIT-licensed (https://www.artany.ai/models/wan-ai). The community has already shipped ControlNet-Wan, ComfyUI-Wan and more; the ecosystem is exploding.Next step
The roadmap promises a 4× speed drop in August, cutting VRAM by another 30 %. By this time next year, your phone may be rendering 1080 p films in real time.
From 8 GB of VRAM to an 8 K screen, Wan AI 2.2 reduces “making a movie” to typing a single sentence. Hollywood isn’t out of reach—it’s already idling in your graphics card, one click away.
Top comments (0)