WantVideo is a third-party aggregator that resells Alibaba's Wan, Google's Veo and ByteDance's Seedance and Seedream video and image models through a single account and a shared credit pool. It runs in the browser with API access, and its pricing page currently states that Stripe payments have been discontinued.
Used this tool? Rate it
Used this tool? Rate it
WantVideo, at wantvideo.ai, is a browser-based platform that lets you generate video and images from several different AI models through one account and one shared credit balance. Its own terms describe it as an all-in-one AI image and video generation platform, reachable through the website, an API, or AI agent integrations such as OpenClaw, Claude Code, and Codex.
The single most important thing to understand before reading any further is what the product actually is, because the marketing does not make it obvious. WantVideo does not build models. It is an AI video aggregator — a third-party reseller of access to other companies' models — Alibaba's Wan family, Google's Veo, and ByteDance's Seedance and Seedream. The site's own footer says this plainly, and it is worth quoting because the homepage copy pulls in the opposite direction: WantVideo is an independent platform and is not affiliated with, endorsed by, or associated with Alibaba, Wan, Google, Veo, ByteDance, or any other model provider.
That disclosure matters, and it is to the operator's credit that it exists. The homepage reads very differently. Scroll it and you will find the phrase "Wan 2.7" repeated dozens of times, attached to nearly every feature heading, in a way that reads as though the platform built the model. It did not. Alibaba did, a fact WantVideo does acknowledge — but in a pricing-page FAQ answer rather than anywhere near the hero copy.
Here the evidence runs thin, and the honest answer is that we do not know. The Terms of Service, effective 26 March 2026, open by saying the Service is "operated by WantVideo" and never name a company. There is no Pte Ltd, no LLC, no incorporation number, no registered address. The footer copyright reads simply "© 2026 WantVideo All Rights Reserved." The only jurisdictional signal anywhere is the governing-law clause, which specifies the laws of Singapore with disputes resolved in Singapore courts — but with no named entity, that clause does not tell you who you would actually be suing. The sole contact channel published anywhere on the site is the email address support@wantvideo.ai.
This is not disqualifying on its own; plenty of small operators run this way. But it should be weighed alongside the payment situation described below, and it is the reason this page is more cautious than a typical product write-up.
A note on names, because the confusion is easy and consequential. Alibaba's own Wan platform lives at a different address entirely. WantVideo at wantvideo.ai is a separate business that resells that model alongside Google's and ByteDance's. If you are looking for the first-party service, this is not it; if you are comparing prices, you are comparing a reseller's markup against a vendor's direct rate.
Being a multi-model platform is the product's actual value proposition, and it is a real one. Rather than holding separate subscriptions with Alibaba, Google, and ByteDance, you hold one balance here and spend it on whichever model suits the shot. The pricing FAQ states it directly: credits work across every model on WantVideo — Wan 2.7, Seedance 2.0, Seedance 1.5, Veo 3.1, and Seedream 4.5 — one credit balance, all models, with the cost per generation shown upfront.
The full model roster, enumerable from the site's own routes, runs to ten entries: wan-2-7, wan-2-7-image, wan-2-6, wan-2-5, wan-2-2-lora, seedance-2-0, seedance-1-5, seedream-4-5, veo-3-1, and nano-banana-2. Three upstream vendors, one checkout.
The platform even coaches you on cost trade-offs, which is more candid than most: a banner at the top of the homepage suggests that for faster and more budget-friendly results you try WAN 2.6, WAN 2.5, or Seedance 1.5, with direct links to each.
Here is where a careful reader needs to slow down, because the site contradicts itself.
The homepage claims Wan 2.7 delivers 4K resolution and 30-second clips with native audio and lip sync. It repeats this in several places: that Wan 2.7 outputs 4K resolution video, the highest native resolution from any Wan AI model; that it generates up to 30 seconds of continuous video in a single generation; that it supports 5 simultaneous video references for multi-character consistency.
The platform's own model page for that same model says something different. The description on /models/wan-2-7 reads: text-to-video, image-to-video, and reference-to-video at up to 1080p and 15 seconds. Same model, same site, two official pages, a fourfold gap in resolution and a twofold gap in duration.
Third-party evidence sides with the model page. WaveSpeedAI, an inference provider that serves Alibaba's Wan 2.7 API directly, lists every video endpoint at 720p/1080p: image-to-video converts images into videos at 720p/1080p, video-edit supports 720p/1080p output, and video-extend supports 720p/1080p output. The 4K figure in that same catalogue belongs to the image endpoints, which output up to 4K resolution with Pro tiers.
An independent guide to Wan 2.7 draws the same line: 4K image output means Wan 2.7 Image supports image generation up to 4K resolution, useful for high-resolution campaign assets, product visuals, and thumbnails. Its video section discusses native audio sync, first and last frame control, multi-reference consistency, and instruction-based editing — never resolution.
So the fair reading is not that the 4K claim is invented. It is that 4K is real on the image side and has been carried across into video marketing where it does not belong. Plan for 1080p video.
Not everything here needs a caveat. Native audio is genuine: the independent guide confirms that Wan 2.7 supports native audio synchronization during generation, matching WantVideo's claim that audio is produced alongside the video rather than layered afterward. First-and-last-frame control, multi-reference consistency, and instruction-based editing are all confirmed upstream capabilities.
The 9-Grid Image-to-Video mode is a genuinely useful framing: upload a 3x3 grid of stills and the model interpolates them into continuous motion, which suits storyboards, comics, and sequential narratives.
The image side is a real second line rather than an afterthought. Three of the ten model routes are image-only, and the pricing tiers quote image counts separately from video seconds. Nano Banana 2 is presented as generating high-resolution AI images up to 4K with text-to-image and image editing. Seedream 4.5 is described as covering text-to-image, multi-image editing, and poster design with readable text rendering.
Because the models differ substantially, the site's own per-model descriptions are the most reliable spec sheet available. WAN 2.6 is listed at up to 1080p and 15 seconds, covering text-to-video, image-to-video, reference-to-video, and video extend, with Standard and Flash modes. WAN 2.2 LoRA is listed at up to 720p and 8 seconds with custom LoRA model support. Seedance 2.0 is described as native audio-video joint generation at 2K with physics-aware motion and director-level camera control, and the page claims a number-one position on the Artificial Analysis leaderboard — a vendor-relayed claim this page does not independently verify.
The strongest case for an aggregator is comparison. If you do not yet know whether Wan, Veo, or Seedance handles your particular kind of shot best, holding one balance and running the same prompt through three engines is cheaper and faster than opening three accounts. The platform's cost guidance makes this easy to do deliberately rather than expensively.
Image-to-video with a 9:16 aspect ratio is the bread-and-butter case. Start from a product photo or a character render, add motion, get a vertical clip. The platform explicitly positions output for YouTube, TikTok, and Instagram.
The 9-Grid mode is unusual enough to be worth naming as its own use case. Illustrators and comic artists with a page of panels can feed the grid in and get interpolated motion out, which is a different workflow from prompting a scene from scratch.
Native audio with synchronized lip movement is the feature that separates this generation of models from the silent-clip era. If your output needs a character to speak, this is the relevant capability — and it is one of the claims that survives independent verification.
The API and agent integrations open the door to batch work and pipeline use. Read the licensing constraints in the Limitations section first, because they are stricter than they look.
Anyone who needs a contractually identifiable counterparty, a stable long-term platform commitment, or verified 4K video delivery should look elsewhere. The reasons are set out below and they are not minor.
1. Create an account. Sign-up supports Google OAuth; the privacy policy notes that account creation collects your name, email address, and authentication details. Every purchase button on the pricing page reads "Sign in to purchase," so an account precedes any transaction.
2. Claim your free credits. The pricing FAQ says that signing up grants free credits, described as enough for your first generation. Free accounts have the same features as paid ones: Wan 2.7 access, native audio, and all supported models. The exact number of free credits is not published, so treat the first run as the way to find out.
3. Pick a model deliberately, not by default. This is the step most users skip and the one that most affects cost. Wan 2.7 is the flagship and the most expensive; Wan 2.6, Wan 2.5, and Seedance 1.5 are the platform's own suggestions for faster, cheaper output. For images, Nano Banana 2 or Seedream 4.5.
4. Write the prompt and attach references. Describe the scene, motion, and camera angles. Upload reference images or videos where character consistency matters — up to five simultaneous video references on Wan 2.7. Toggle first-and-last-frame control if you need to pin the start and end points.
5. Set resolution, duration, and aspect ratio. Duration runs from 5 to 30 seconds; aspect ratios are 16:9, 9:16, and 1:1. On resolution, set expectations at 1080p rather than 4K for video, for the reasons documented above.
6. Check the credit cost before you commit. The platform shows the credit cost of each job before you press Generate. Published reference points: roughly 150 credits for a 5-second 480p video, roughly 600 credits for a 10-second 720p video, with 4K and 30-second clips costing more. Reading that number before each run is the single most effective cost control available.
7. Download and check the licence tier. Output arrives with native audio and no watermark on paid plans. Before using anything commercially, confirm which plan you are on — commercial licensing is not included at every tier.
Creators comparing models get the clearest benefit. One balance, ten models, no separate signups.
Short-form social producers working in 9:16 from stills, at volume, where 1080p is entirely sufficient and the per-clip cost matters more than maximum fidelity.
Storyboard-driven illustrators who can exploit the 9-Grid mode.
Developers doing batch generation via the API — subject to the redistribution restrictions.
Hobbyists and first-time users, who can test the whole roster on free credits without committing money. Given the payment notice, this is arguably the most sensible way to approach the platform right now.
Not for: teams that require a named contractual counterparty; anyone with a hard 4K video deliverable; agencies needing guaranteed service continuity, since the platform's own annual-plan FAQ concedes that upstream changes can affect availability; and developers who want to build a product on top of the API, which the terms forbid.
WantVideo is a web application with two programmatic entry points. The terms name three access routes: the website, the API, and AI agent integrations such as OpenClaw, Claude Code, and Codex. No native mobile or desktop client is mentioned anywhere in the terms or on the site.
Hosting is disclosed in the privacy policy as Railway, with payments historically through Stripe and AI generation through the third-party model providers. Authentication supports Google OAuth.
Output formats cover 16:9 landscape, 9:16 vertical, and 1:1 square, with durations from 5 to 30 seconds, and the platform positions results as ready for YouTube, TikTok, and Instagram.
Generation history is available through a Creation History view, present in the navigation of every model page and at the /generations route.
The pricing page carries a banner that changes how everything below it should be read. It states that payments via Stripe have been discontinued, that the platform is no longer accepting payments through Stripe, that all existing Stripe subscriptions have been cancelled and users will not be charged again, and it directs anyone who notices an unexpected charge to contact support@wantvideo.ai.
The four paid plans remain displayed on that same page beneath the notice, and no alternative payment processor is named anywhere on the site.
Compounding this, the legal documents have not been updated to match. The Terms of Service, effective 26 March 2026, still state that all payments are processed through Stripe. The Privacy Policy, effective 14 April 2026, still lists Stripe as the payment processor and as a trusted service provider. Both documents post-date or nearly coincide with the notice and neither reflects it. These three sources are reported side by side here without attempting to reconcile them, because reconciling them would mean guessing.
The practical reading: treat the prices below as published rates whose current purchasability you must confirm directly with the operator before planning around them.
All four are quoted at an advertised 50% discount on annual billing.
Mini — $9.99/month, listed against $19.9, or $119.9/year. 15.6K credits per year, $0.77 per 100 credits. Up to 4,100 seconds of video and 1,560 images.
Starter — $14.99/month, listed against $29.9, or $179.9/year. 24K credits per year, $0.75 per 100 credits. Up to 6,400 seconds and 2,400 images.
Creator — $24.99/month, listed against $49.9, or $299.9/year. 48K credits per year, $0.62 per 100 credits. Up to 12,800 seconds and 4,800 images. Marked "Popular."
Pro — $49.99/month, listed against $99.9, or $599.9/year. 120K credits per year, $0.50 per 100 credits. Up to 32,000 seconds and 12,000 images.
Every tier includes access to all models, native audio and lip sync, no watermark, and a no-AI-training commitment. Commercial licence and priority support appear only on Creator and Pro.
There is also a one-time Credit Pack of 20,000 credits with no expiration and no subscription required.
Cost scales with resolution and duration. The published reference points are approximately 150 credits for a 5-second 480p video and approximately 600 credits for a 10-second 720p video, with 4K and 30-second output costing more. Each job displays its cost before generation.
Three rules deserve attention. Subscription credits expire 30 days after issuance and unused ones do not roll over, while one-time credits never expire. Refunds must be requested within 30 days of purchase, cover only unused credits at the original purchase price, and are processed within 5 to 10 business days. And unused credits on terminated accounts are forfeited, unless the termination was initiated by the platform without cause.
One line in the annual-plan FAQ is more revealing than its placement suggests: you receive all credits upfront at the start of the year, and if upstream changes affect availability, remaining time is refunded pro-rata. That is the operator acknowledging that service continuity depends on model providers it does not control.
Signing up grants free credits, characterised as enough for your first generation, with the same feature set as paid accounts. The quantity is not published.
Going direct to the model vendors is the most obvious alternative and the one most worth considering here. Alibaba, Google, and ByteDance all sell access to these models themselves. You lose the single-balance convenience and gain a named counterparty, first-party support, and no reseller margin.
Other multi-model aggregators occupy the same niche. Several inference providers list the identical Alibaba Wan endpoints — WaveSpeedAI, cited elsewhere on this page for its published specs, is one such catalogue. If the aggregation model appeals but this particular operator's opacity does not, that is where to look.
Full-featured video suites such as invideo, Runway, or Pika bundle generation with editing, timeline work, and asset management. They cost more and do more; a pure generation front-end like this one is the right choice only if you already have an editing workflow elsewhere.
Open weights run locally is viable for the Wan family specifically, since Wan models have a substantial open-source presence. This trades money for GPU time and setup effort, and removes both the reseller and the payment question entirely.
The honest framing for comparison: what you are buying here is convenience and price arbitrage across three vendors, not capability you cannot get elsewhere. Every model on this platform is available from its maker. Weigh the convenience against the identity and continuity questions below.
No named legal entity. The terms, the privacy policy, and the footer all identify the operator only as "WantVideo." No company form, no registration number, no address. Singapore governing law is specified without a Singapore entity to attach it to. If a dispute arose, the counterparty you would name is not published.
Stripe payments discontinued, with paid plans still displayed. Documented in full above. No replacement processor is named.
The legal documents contradict the payment notice. Terms and privacy policy both still describe Stripe as the active processor, despite post-dating or coinciding with the notice. Stale legal text is a maintenance signal worth weighing.
The 4K video claim is not supported. The homepage says 4K and 30 seconds; the platform's own model page says up to 1080p and 15 seconds; the upstream API catalogue lists 720p/1080p for every video endpoint. The 4K figure belongs to the image models.
Generated-content ownership is conditional. You own your output subject to the terms of the underlying AI models — meaning the operative licence is upstream, and you are responsible for complying with it.
Commercial licensing is tier-gated. Not available on Mini or Starter.
Subscription credits expire in 30 days and do not roll over. Unused allowance is lost each cycle.
Credits are forfeited on termination. Unless the platform terminates without cause.
API redistribution and competing products are prohibited. The terms bar redistributing API access or building competing services on the API, and separately bar circumventing usage limits or reselling API access. This closes off most reseller and wrapper business models.
Upstream availability risk is acknowledged by the vendor. The pro-rata refund language in the annual-plan FAQ is an admission that the roster can change out from under a subscriber.
Broad disclaimers and unilateral change rights. Service is provided "as is" and "as available" with no warranties, no guarantee of uninterrupted or error-free operation, and no guarantee that output meets expectations. The platform may modify, suspend, or discontinue any part of the service at any time, may change pricing at any time, and may remove content and suspend accounts without prior notice.
No independent user reviews exist. No Trustpilot profile was found for the domain, and a search for independent reviews returned only category round-ups with no coverage of this platform specifically. Combined with the absence of a named operator and the payment situation, this absence of third-party signal is itself worth weighing rather than treating as neutral.
Advertising data sharing, with a nuance. The privacy policy uses Microsoft Advertising and its UET tag, which may collect IP address, pages visited, and ad interaction data. Its CCPA section states that personal information is not sold, then states that limited data is shared with advertising partners for cross-context behavioral advertising. Under CCPA those are distinct concepts; both statements are reproduced here rather than merged. Opt-out routes are provided.
Directory metadata is out of date. The Toolso listing for this tool carries titles in all ten languages containing the word "Spicy" or its localized equivalents, implying adult-oriented positioning. A word-count over the current homepage HTML and all three legal documents returned zero occurrences of spicy, NSFW, adult content, uncensored, or unfiltered. The current terms in fact prohibit sexually explicit content involving minors, and the compliance notice restricts users to their own likeness or original virtual faces. The listing description also names "Wan 2.2 Animate," which does not appear in the current model roster — the actual entry is wan-2-2-lora. The current site should be taken as authoritative over that metadata.
Age limit and data transfers. The service is not intended for children under 13. Information may be transferred to and processed in countries other than the user's own, with continued use constituting consent.
The clearest commitment here is also the most relevant to creators: the platform states it does not use prompts, uploaded images, or generated content to train AI models, adding that creative inputs and outputs belong to the user. The same commitment appears in the terms and is listed as "No AI training" on all four pricing tiers. Repetition across three surfaces with consistent wording is a reasonable indicator of intent.
Data collected covers account information, payment details received from the processor (full card numbers are not stored), usage data including prompts and generation parameters, credit and transaction history, API request logs, and device data such as IP address, browser, operating system, and referring URLs.
Prompts and reference images are transmitted to the third-party model providers to fulfil requests — an unavoidable consequence of the aggregation model, and disclosed rather than hidden.
The licence granted to the platform over user content is narrowly scoped: limited, worldwide, non-exclusive, royalty-free, solely as necessary to provide and improve the service.
Users may request access, correction, deletion, marketing opt-out, and advertising opt-out by email, with a 30-day response commitment. California residents have the additional CCPA rights described above. Security is stated as industry-standard with HTTPS in transit, alongside the customary acknowledgment that no internet transmission is entirely secure.
No. It is an aggregator that resells access to models built by others — Alibaba's Wan family, Google's Veo, and ByteDance's Seedance and Seedream. Its own footer states that it is an independent platform not affiliated with, endorsed by, or associated with any of those providers, and its pricing FAQ acknowledges that Wan 2.7 is the latest AI video model from Alibaba.
No. Alibaba's first-party Wan service runs on a different domain entirely. WantVideo at wantvideo.ai is a separate reseller that offers Wan alongside Google's and ByteDance's models. If you want the first-party service, this is not it.
This could not be established. The terms say only that the service is "operated by WantVideo," with no company form, registration number, or address anywhere in the terms, the privacy policy, or the footer. Governing law is Singapore and disputes go to Singapore courts, but no Singapore entity is named. The only published contact is support@wantvideo.ai.
Check with the operator before assuming so. The pricing page states that payments via Stripe have been discontinued, that the platform no longer accepts payments through Stripe, and that existing Stripe subscriptions have been cancelled. No replacement processor is named, while the four paid plans remain displayed on the same page. The terms and privacy policy still describe Stripe as the active processor, so the documentation is internally inconsistent on this point.
The evidence says plan for 1080p. The homepage claims 4K and 30-second clips, but the platform's own model page for the same model says up to 1080p and 15 seconds, and the upstream API catalogue lists every Wan 2.7 video endpoint at 720p/1080p. The genuine 4K capability belongs to the image models, and appears to have been carried across into video marketing where it does not apply.
One balance spends across every model, with each job's cost shown before you generate. Roughly 150 credits buys a 5-second 480p clip and roughly 600 buys a 10-second 720p clip. Expiry depends on origin: subscription credits expire 30 days after issuance and do not roll over, while one-time credit-pack credits never expire. Credits on a terminated account are forfeited unless the platform terminated without cause.
Only on the Creator and Pro plans, which list a commercial licence; Mini and Starter do not. And there is a second condition that applies at every tier: the terms grant you ownership of output subject to the terms of the underlying AI models, so the binding limits are set by Alibaba, Google, or ByteDance and you are responsible for complying with them.
The platform says no, in three places with consistent wording: the terms state it does not use inputs or generated content to train AI models, the privacy policy states that prompts, uploaded images, and generated content are not used for training, and every pricing tier lists "No AI training." Note separately that prompts and reference images are transmitted to the upstream model providers to fulfil generation requests, whose own policies govern that leg.
No. The terms prohibit redistributing API access and prohibit building competing services using the API, and separately bar reselling API access or circumventing usage limits. The API is for using the service, not for wrapping it.
None was found. There is no Trustpilot profile for the domain, and searches for independent reviews returned only broad category round-ups that do not cover this platform. Accordingly, this page cites no ratings at all and rests its third-party evidence on verifiable upstream model specifications instead.
