Wan 2.1
Wan 2.1 by Alibaba: an open video model for text to video and image to video, in a small 1.3B and a large 14B size, with many add-ons.
What it is
Released in 2025, Wan 2.1 made good open video generation practical. The 14B models come in 480p and 720p versions for text to video (t2v) and image to video (i2v); the 1.3B model runs on small cards. A large family of add-ons builds on it: VACE, first-last-frame (FLF2V), Fun Control and InP, InfiniteTalk and MultiTalk, FantasyTalking, UniAnimate, ReCamMaster, speed-up LoRAs.
Files and nodes
Load Diffusion Model (wan2.1_t2v_14B, wan2.1_i2v_480p_14B and others); Load CLIP with umt5_xxl (type wan); Load VAE with wan_2.1_vae; for image to video also Load CLIP Vision with clip_vision_h. EmptyHunyuanLatentVideo or WanImageToVideo; ModelSamplingSD3; one KSampler.
Settings that work
832×480 (480p) or 1280×720 (720p); 81 frames at 16 fps; about 20 to 30 steps; cfg about 5 to 6 in the official templates (lower with speed-up LoRAs: 4 to 8 steps, cfg 1); shift about 5 to 8. Prompt: detailed sentences in English or Chinese.
What fits
LoRAs for Wan 2.1 and for the same size: 1.3B and 14B are not interchangeable. Many 14B LoRAs also work on Wan 2.2 14B.
How the app helps
Shown as Wan 2.1 on model cards. The InfiniteTalk, FantasyTalking, UniAnimate and ReCamMaster lessons use Wan 2.1 models.