Quick Verdict
StepFun Video, commonly called Jieyue Video in Chinese, is StepFun’s hosted AI video product for ordinary creators. Its canonical user surface is yuewen.cn/videos. It turns a written description or an uploaded image into a short video candidate. It is not the same artifact as the Step-Video-T2V or Step-Video-TI2V model repositories. The product manages accounts, credits, job settings, moderation, and downloads; repositories contain particular model code, weights, licenses, and deployment instructions. A repository parameter count or benchmark cannot be copied directly into a claim about the current web mode.
As of July 21, 2026, product materials still frame the tool around clips up to roughly ten seconds. Different modes, references, resolutions, campaigns, or accounts may expose shorter choices, so the job screen remains authoritative. The old “free beta” label no longer describes the product safely. Do not assume permanent unlimited generation, and do not invent per-job credit or cash prices when the live account controls them.
Best For
StepFun Video is best for short-video and animation creators, social teams, ad designers, teachers, and independent filmmakers who need to preview a shot before editing. It is particularly useful for answering a narrow visual question: how should this character image move, or does this prompt produce a usable camera idea? It should feed an editor rather than replace one.
Projects requiring multi-minute continuity, stable acting, exact lip sync, complex crowds, frame-level animation, or guaranteed enterprise throughput need a broader stack. Developers evaluating Step-Video repositories should separately review model cards, licenses, GPU requirements, inference time, and safety controls. Web credits do not fund self-hosting, and repository features do not automatically appear in the consumer product.
Key Features
- Text-to-video: describes subject, action, environment, visual style, camera movement, and pacing in Chinese to create one short shot.
- Image-to-video: uses an authorized image as a visual foundation for character, product, illustration, or scene motion.
- Short-clip options: the product presents currently available duration, aspect-ratio, and mode choices; “up to about ten seconds” does not mean every mode offers ten seconds.
- Credit-based jobs: account credits or allowances are consumed according to the confirmation screen. Longer or higher-specification jobs may use different amounts.
- Separate model ecosystem: StepFun also publishes Step-Video research and open model projects, whose local output, licensing, and hardware requirements are distinct from the web service.
Use Cases
Break a long script into single shots. Each prompt should define the subject, start and end of the action, shot size, camera movement, environment, and lighting. For image-to-video, use images you created, licensed, or received written permission to process. Keep evidence for photography, character IP, logos, client assets, and publication scope. Test motion with a shorter or lower-consumption setting before spending credits on the longest available option.
A real person’s image or voice requires informed, purpose-specific consent. Public figures are not public-domain source material and must not be placed in false endorsements, impersonations, or misleading events. Minors require guardian consent and stronger safeguards against dangerous imitation, humiliation, sexualization, and deceptive commercial contexts.
Inspect identity, hands, clothing, products, logos, text, and motion frame by frame. Clearly label synthetic content, preserve required visible or invisible marks, and do not crop or re-encode simply to bypass a watermark. The destination platform may impose additional rules for AI media, advertising, music, and likenesses. Generated historical, news, educational, or product claims also require factual review.
Pricing
StepFun Video uses account allowances or credits. Complimentary credits, per-mode consumption, expiration, purchases, failed-job refunds, and campaign rewards may change. This guide therefore does not publish an unsupported credit number. The job confirmation screen should show the selected mode, duration, and current consumption before submission. If it does not, run a low-cost test and record the balance change. The ability to make a clip up to about ten seconds is a capability boundary, not a promise that every account receives that specification free.
Calculate cost per usable shot, not per generation. A single approved clip may require multiple samples, plus queue time, failed attempts, repair, editing, and licensed music. Before purchasing credits for a commercial campaign, confirm the current commercial-use permission, storage and download period, renewal behavior, and refund rules. Payment buys service access; it does not guarantee exclusive output, copyright registration, or freedom from protected third-party elements.
Pros
- Canonical official surface for low-friction Chinese text-to-video and image-to-video experiments.
- A roughly ten-second upper range can fit storyboards, social assets, and visual concept testing.
- Credit-based online jobs avoid installing and operating a large local model for initial evaluation.
- A separate Step-Video research ecosystem gives capable engineering teams another route to investigate.
Cons
- Credit grants, expiration, and per-job consumption are dynamic, making the old free-beta description unreliable.
- “Up to ten seconds” is not a guaranteed option for every mode, account, input type, or resolution.
- The hosted product and open model projects are easy to conflate even though features, licenses, hardware, and output terms differ.
- Character continuity, small text, complex motion, and physical relationships can fail and require sampling and repair.
Alternatives
| Tool | Best for | Main advantage | Main limitation |
|---|---|---|---|
| Kling AI | Broad Chinese video workflows | Mature online modes | Quotas and usable-shot cost require testing |
| Vidu | Subject references and recurring characters | Multi-image references, subjects, endpoint frames | Credit expiration and consistency need management |
| Zhipu Qingying | Creators drafting in Zhipu Qingyan | Connected Chinese chat and video workflow | Limited professional editing and long continuity |
| Tencent Hunyuan Video | Teams comparing web, API, and self-hosting | Multiple official delivery routes | Their pricing and licenses are easy to conflate |
| Jimeng AI | Integrated image, video, and template creation | Convenient Chinese creative ecosystem | Account benefits and platform policies can change |
FAQ
What is the canonical StepFun Video URL?
Use https://yuewen.cn/videos. The consumer product is StepFun Video; Step-Video model repositories are separate engineering artifacts.
How long can StepFun Video generate?
Current product material describes clips up to about ten seconds, but a mode, reference type, aspect ratio, or account may expose other durations. Check the job screen.
How many credits does one video cost?
Consumption depends on the current mode, duration, specification, and campaign. Use the confirmation screen and observed balance instead of a stale third-party number.
How does the product differ from Step-Video open models?
The product is hosted and account-based. The model projects provide code and weights. Product terms, credits, model licenses, hardware cost, and available outputs must be reviewed separately.
Can videos made with paid credits be used commercially?
Payment alone is not enough to answer. Check the current terms and secure rights to every source image, person, character, music track, trademark, and client asset. Output rights are not guaranteed.
Must I preserve watermarks and disclose AI generation?
Follow legal, StepFun, and destination-platform requirements for visible and invisible identifiers. Do not evade a mark by cropping or transcoding, and fact-check any realistic claim.
Bottom Line
Evaluate StepFun Video as a credit-based hosted product for creating short candidates up to roughly ten seconds, not as a showcase page for Step-Video model parameters. Test five representative shots and record duration choices, credits, queue time, usable rate, and repair work. Compare the result with Vidu or Kling. Before commercial publication, complete source licensing, likeness and guardian consent, synthetic disclosure, watermark and platform checks, factual review, and an output-rights assessment.