{"id":"kandinsky-s2v-1-0-lite","slug":"kandinsky-s2v-1-0-lite","name":"Kandinsky S2V 1.0 Lite","orgId":"kandinsky","family":"kandinsky-s2v-1-0-lite","releaseDate":null,"availability":"ga","openWeights":true,"license":"MIT (code and weights)","paramsTotal":null,"paramsSource":null,"contextWindow":null,"knowledgeCutoff":null,"modelCardUrl":"https://arxiv.org/abs/2610.19290","hfRepo":null,"modalities":["video"],"aliases":["Kandinsky S2V 1.0 Lite","kandinsky-s2v-1-0-lite","kandinsky-s2v-1.0-lite","Kandinsky-S2V-1.0-Lite","kandinskylab/Kandinsky-S2V-1.0-Lite","Kandinsky S2V Lite"],"aliasNote":null,"origin":"tracked","tracked":{"trackedSince":"2026-11-04","announced":"2026-10-13","blurb":"The 2B size of Kandinsky Lab's Kandinsky S2V 1.0 speech-to-video family: it animates a person from a single reference image, a speech track and a text prompt, with lip motion, expression, pose and gesture following the speech. It does not generate audio. Output is 25 fps in SD, with long speech generated in 5-second chunks. Initialized from Kandinsky 5.0 Video. Code and weights under MIT.","api":"none","openWeights":"yes","params":{"text":"2B","source":"Kandinsky Lab's paper and model card (2.3B DiT backbone)"},"facts":[{"label":"Task","value":"Speech-to-video (audio-driven avatar animation) from one reference image, a speech waveform and a text prompt; audio is a condition only"},{"label":"Output","value":"25 fps video in SD at the reference frame's aspect ratio; long speech in 5-second chunks"},{"label":"Architecture","value":"3D Diffusion Transformer with flow matching in a frozen Hunyuan video VAE latent; Qwen2.5-Omni text and audio front end; three-scale FramePack visual memory"},{"label":"Training data","value":"1.17M single-speaker clips, captions in English and Russian"},{"label":"Checkpoints","value":"Base, pi-Flow distilled (16 steps instead of 50) and Diffusers checkpoints on Hugging Face; demo Space"},{"label":"License","value":"MIT (code and weights)"},{"label":"Related models","value":"Separate from Kandinsky S2V 1.0 Pro and from Kandinsky 6.0 Video"}],"claims":["Lab claim: in a side-by-side human evaluation, Kandinsky S2V 1.0 Pro performs comparably to or better than the strongest open-source baseline and two proprietary commercial models on average, and is preferred for lip sync.","Lab claim: the distilled Pro model runs in 16 steps without classifier-free guidance, a 5.4x end-to-end speedup over the base model."],"sources":[{"kind":"paper","title":"Kandinsky S2V 1.0: Audio-Driven Avatar Video Generation (arXiv)","url":"https://arxiv.org/abs/2610.19290","date":"2026-10-13"},{"kind":"repo","title":"kandinskylab/Kandinsky-S2V-1.0-Lite model card (Hugging Face)","url":"https://huggingface.co/kandinskylab/Kandinsky-S2V-1.0-Lite"},{"kind":"repo","title":"kandinskylab/kandinsky-s2v (GitHub)","url":"https://github.com/kandinskylab/kandinsky-s2v"},{"kind":"lab","title":"Kandinsky S2V 1.0 project page","url":"https://kandinskyavatar.github.io/"}],"firstBenchmark":null},"organization":{"id":"kandinsky","name":"Kandinsky","slug":"kandinsky","country":null,"url":"https://kandinskylab.ai","color":"#A39E8F","mark":"KA"},"price":null,"answer":{"word":"Pending benchmarks.","scope":"Announced Oct 13, 2026; no independent results yet.","lead":"As of Nov 22, 2026, Kandinsky S2V 1.0 Lite is pending benchmarks on isitsota: announced Oct 13, 2026, with no independent results on the boards we track yet.","confidence":"low"},"status":"PENDING_BENCHMARKS","verdicts":[]}