930 B
Wan is Alibaba Tongyi Lab's open-source video generation family (Apache 2.0 license) covering text-to-video, image-to-video, speech-to-video, motion control, and video editing. Wan 2.1 offers T2V/I2V in 1.3B (8GB VRAM) and 14B (24GB+) sizes at 480p/720p, plus Fun variants for camera control, inpainting, and depth/pose/canny conditioning. VACE provides unified video editing with reference-to-video, video-to-video, and masked editing. Wan 2.2 introduces MoE architecture for higher quality at same compute cost, TI2V-5B hybrid generation (runs on 4090), and S2V for speech-driven video with lip-sync. Wan 2.2 Animate handles character animation and subject replacement with relighting LoRA. Wan Move offers point-level motion control via dense trajectory guidance. Key strengths: high temporal consistency, bilingual prompts, multiple aspect ratios at 24fps. Parameters: 20-50 steps, cfg_scale 3-7, typically 81 frames (~3.4s).