Wan 2.2 suits teams that want to animate existing art while controlling the generation pipeline locally. Image-to-video gives a scene a defined starting image, though it cannot guarantee that a character stays consistent throughout the clip. For adult character work, inspect faces, bodies and clothing across the whole take. Choose the variant before budgeting hardware: the A14B models and TI2V-5B have very different memory needs. Start with an official inference configuration, then compare any quantized or accelerated setup against that baseline. ComfyUI and Diffusers support offer practical integration options. Use local prompt processing as well as local inference if source-art privacy matters, and budget time for review and rejected takes.
Commercial use & license
The official models are released under Apache 2.0. Keep the license and required notices when redistributing the models or modified versions; check separate terms for custom checkpoints, adapters, source art and services.
Adult content rules
The repository warns against unlawful or harmful sharing, malicious disclosure of personal data, misinformation and targeting vulnerable people. For adult work, also check input rights, applicable law and the rules of any hosting or distribution service.
What you need
The official 720p single-GPU examples specify at least 80 GB VRAM for T2V-A14B and I2V-A14B, and 24 GB for TI2V-5B with model offloading, dtype conversion and CPU text encoding. A14B supports 480p/720p; TI2V-5B supports 720p at 24 fps, using 1280×704 or 704×1280. Include the required encoder and VAE files.
Ranks are recommendations for the task described, based on reviewed documentation. Model ranks cover non-explicit development work. We have not run practical benchmarks. Check content policies and commercial terms separately. How the ranks work →