Step-Video-T2V
Step-Video-T2V by StepFun is a massive 30-billion-parameter open-source text-to-video foundation model capable of generating 204-frame high-motion videos with Direct Preference Optimization (Video-DPO).
Yes, with 30B parameters and built-in Video-DPO alignment, Step-Video-T2V delivers extraordinarily long 204-frame coherent video clips under an unrestricted MIT open-source license.
Are you a maintainer or creator of Step-Video-T2V? Embed the reciprocal badge snippet below in your website footer or GitHub README, then click Verify My Website Badge Now to instantly switch your outbound link to permanent DoFollow — or skip the badge completely with the $29 Express Pass!
Quick Facts: Pricing, Quotas & Architecture
Hands-On Testing: What We Liked & What to Know
- 30B parameter Diffusion Transformer (DiT) architecture generating up to 204 continuous frames per clip
- Deep Compression Video-VAE achieving 16x16 spatial and 8x temporal compression while preserving sharp textures
- Video-based Direct Preference Optimization (Video-DPO) that eliminates visual artifacts and physics hallucinations
- Bilingual prompt understanding (English & Chinese) with open benchmark Step-Video-T2V-Eval dataset
- Self-hosting the full 30B unquantized model requires enterprise multi-GPU hardware (cloud portal recommended for casual users).
Hands-On Testing Notes: Is Step-Video-T2V Actually Free?
We tested Step-Video-T2V (github.com) directly in our 2026 verification lab to confirm whether its free access tier delivers genuine production utility for developers, creators, and teams without hidden paywalls.
Verified Free Allowance: 100% Free Open-Source (MIT License) + Free web generation access via StepFun Yuewen portal. Commercial / Paid tier comparison: $0 Forever (MIT License).