TellWorldAI World Archive

Home›News›Release

ReleaseGenerative AI for everyone

StepFun and Geely open-source Step-Video-T2V and Step-Audio

On 17 February 2025 StepFun and Geely jointly open-sourced Step-Video-T2V, a 30-billion-parameter video model, and Step-Audio, a speech model family.

Background StepFun (阶跃星辰) is a Chinese AI company that works on multimodal models. Geely (吉利汽车) is a carmaker, and the two companies released their models together in February 2025 [3].

What happened In February 2025 StepFun and Geely jointly open-sourced two multimodal models: Step-Video-T2V and Step-Audio [3]. On 17 February 2025 StepFun released the weights, inference code, a Turbo variant and the technical report for Step-Video-T2V [1].

Key details Step-Video-T2V has 30 billion parameters and generates videos of up to 204 frames [1]. It was released under the MIT license [1].

On the same day StepFun released Step-Audio. The release included Step-Audio-Chat, which has 130 billion parameters, as well as Step-Audio-TTS-3B and Step-Audio-Tokenizer [2]. TTS stands for text-to-speech, so Step-Audio-TTS-3B converts written text into spoken audio.

StepFun also published two benchmarks: StepEval-Audio-360 for speech and Step-Video-T2V-Eval for video [1][2]. A benchmark is a standard set of test questions used to compare models.

Significance The release was a joint open-sourcing with Geely. It put a video generation model and a speech model into public hands on the same day, and both came with technical reports and benchmarks.

Terms used Text-to-video generation produces a short clip from a written description. A Turbo variant is a version tuned to run in fewer steps. With fewer steps each clip takes less time to generate, usually at some cost in quality.

What the MIT licence allows The MIT licence is a permissive open-source licence. It allows people to use, copy, modify and redistribute the software, including for commercial purposes, as long as the original copyright notice is kept. For a video model, developers can build products on the released weights under these terms.

Follow-up The same base model was extended on 17 March 2025 to Step-Video-TI2V, an image-to-video variant. A single open release therefore led to a second model built on the same foundation within a month.

Sources

  1. 1
  2. 2
    stepfun-ai/Step-AudioGitHub (StepFun)
  3. 3
    StepFunWikipedia

Cite this page: StepFun and Geely open-source Step-Video-T2V and Step-Audio. TellWorld, https://www.tellworld.net/news/step-video-t2v-2025. Updated 2026-11-08

Related news

View all