9 open-source video generation model families to know in 2026, from Wan2.2 to Mochi 1
TheTuringPost · x · 2026-09-27
Turing Post surveys nine open-source video generation model families worth knowing in 2026, comparing quality, speed, hardware needs and licenses across text-to-video, image-to-video, editing and audio workflows.
- Wan2.2: Alibaba's newest downloadable open-weight flagship under Apache 2.0, with a 27B MoE design (separate high/low-noise experts) plus a dense TI2V-5B model running 720p/24fps on a single consumer GPU.
- Wan2.1: the low-VRAM entry point — the 1.3B text-to-video model needs 8.19 GB VRAM for 5-second 480p; suite covers t2v, i2v, editing, t2i and video-to-audio.
- HunyuanVideo 1.5: Tencent's November 2025 rebuild cut parameters from 13B to 8.3B while improving quality and runs on consumer hardware.
- Also covered: LTX-2/2.5, MiniMax H3, CogVideoX, Mochi 1, Kandinsky 5.0, and Stable Video Diffusion.
Notably, later Wan-branded hosted releases have no downloadable repos, so Wan2.2 remains the newest downloadable Wan video model.
Related event: Open-Source Video Generation Models to Watch in 2026(3 posts)→
More from Multimodal
- AI art before it was Deforum: magic from August 2022 — makeitrad1 · 2026-09-27
- Google Flow build now points to Nano Banana 2.1, hinting at iterative image model update — testingcatalog · 2026-09-27
- Google opens video tool vids to everyone, powered by Gemini Omni 1.1 — vista8 · 2026-09-27
- Recreating the viral "Tharog" clip with the same Seedance 2.5 prompt — iamfakhrealam · 2026-09-27
- Grok Imagine agent produces unexpected results — RachelVT42 · 2026-09-27
- New AI Skill routes one brief across multiple image models using Arena rankings — woulatte · 2026-09-27