What is Vidu?
Vidu is a generative media platform from ShengShu AI HK Limited. Its web product and API create video from text, a source image, multiple subject references, start and end frames, or a sequence of frames. The current Q3 family offers Pro, Turbo, Mix, and fast variants across different modes and resolutions. The broader platform includes image generation, sound effects, text to speech, voice cloning, lip sync, motion sync, video extension, upscaling, digital characters, templates, and one-click film workflows.
For creative teams, Vidu's notable feature is reference-driven consistency. Instead of describing the same character or product in every shot, a user can supply reference images and ask the model to preserve subjects and scene attributes. This can reduce drift, but it does not guarantee identity, geometry, logos, packaging, clothing, or continuity. Treat each generation as footage to inspect rather than a locked design asset.
Video modes and practical workflow
Text-to-video is useful for broad motion concepts when no source artwork exists. Image-to-video animates a chosen frame and gives more control over composition. Start/end-frame generation targets a transition between two images. Reference-to-video prioritizes subject consistency, while multi-frame generation creates linked segments from several frames. Vidu Q3 documentation lists clips from one to sixteen seconds for core text, image, and start/end modes, with reference modes beginning at three seconds.
A productive workflow begins with a shot list. Lock aspect ratio, duration, reference set, camera action, and forbidden changes. Generate low-cost tests, select one motion direction, then raise resolution. Export the strongest takes into a conventional editor for cuts, audio, typography, color, cleanup, disclosure, and brand review. Generative text, signs, hands, reflections, small product details, and object interactions deserve frame-by-frame inspection.
Reference images introduce provenance obligations. Confirm copyright, trademark, model release, property rights, and permission for every recognizable person or voice. Do not upload customer assets, unreleased products, children, medical imagery, or confidential locations unless the contract, policy, and legal basis allow that processing.
Credits and API cost
Vidu's consumer page says each user receives 40 free credits per month. Its pricing FAQ distinguishes subscription credits, valid for 30 days, from purchased and bonus credits, documented as valid for two years. Paid subscriptions through Stripe renew automatically, and the page says purchases are generally nonrefundable. The visible public page does not expose every subscription card as crawlable text, so confirm current plan price, watermark, commercial rights, priority, concurrency, and resolution at checkout.
The API is more transparent: one credit costs $0.005. For Q3, normal documented video rates vary from $0.035 per second for Q3 Turbo at 540p to $0.145 per second for Q3 Mix reference video at 1080p. Q3 Pro spans $0.045 to $0.12 per second across 540p–1080p; Q3 Turbo spans $0.035 to $0.065. Eligible off-peak rates are often roughly half, but Mix does not list off-peak support. Audio, upscaling, lip sync, multi-frame segments, prompt recommendations, and voice cloning have separate charges.
The price of a usable clip is the generation rate multiplied by attempts, not the cost of one successful render. Measure acceptance rate by shot type, average attempts, queue time, post-production minutes, and discarded-credit cost. API integrations should set per-user budgets, concurrency limits, idempotency, task cancellation, moderation handling, and alerts.
Commercial use, disclosure, and output review
Vidu's API terms state that the service does not restrict commercial use. The same terms require users to hold the necessary rights, notices, consents, and permissions for input processing and end users. They warn that outputs may be inaccurate or offensive, require human evaluation, prohibit using output for decisions with legal or material effects on a person, and forbid removing or falsifying marks that identify deep-synthesis output.
Commercial permission is not an indemnity or a copyright guarantee. A generated character may resemble a public figure; a logo may mutate into a protected mark; music or imagery can raise third-party claims. Keep input records, prompt and model versions, human approvals, licenses, disclosure labels, and final edits. For advertising, confirm product attributes and legal claims independently.
Privacy and data location
The privacy policy effective April 29, 2026 covers the website, API, and related products. It says Vidu collects account information and personal data embedded in text, images, video, feedback, and user content. It may act as controller for personal use and processor for commercial customers. Customers integrating the API remain responsible for end-user notice, authorization, data minimization, and rights requests.
The policy says personal data may be processed across borders and acknowledges transfer and storage in Singapore. Users have access, correction, deletion, objection, restriction, and portability rights where applicable. Before enterprise use, obtain the DPA, subprocessor list, retention and backup schedule, deletion behavior for inputs and outputs, support-access controls, incident terms, and model-training policy; the public excerpts do not answer every one of these questions.
Verdict
Vidu is a capable option for short reference-driven video and programmatic generation, with unusually detailed API price tables. Its value depends on consistency and acceptance rate in the exact shot types a team needs. Run a controlled benchmark using owned references, calculate cost per approved second, inspect rights and disclosure rules, and test deletion and moderation. Use Vidu as a shot-generation layer inside a reviewed production pipeline—not as an automatic final-publish system.
