Vidu Q4 Preview: AI Video Benchmark Rank, Pricing, and Guide

👁 121 reads

Article updated continuously. Current as of October 8, 2026, 2:00 PM UTC.

Vidu Q4 Preview is the newest video model from Beijing-based ShengShu Technology, released on October 7, 2026. Just a day later, it took 3rd place on the Artificial Analysis Image-to-Video leaderboard with an Elo rating of 1179-just 2 points behind MiniMax H3 and 16 points behind the category leader, MiniMax H3 Max. For comparison, Vidu’s previous model, Q3 Pro, ranked 19th. The model produces 3-16 second clips at up to 4K resolution with native audio, costing $0.12 per second for 1080p via the API, which works out to $7.20 per minute.

What is Vidu Q4 Preview?

Vidu is a video generation engine developed by ShengShu Technology, a team founded by researchers from Tsinghua University in Beijing. Following Q1, Q2, and Q3, the preview of Q4 rolled out on October 7, 2026. The developers position it as a “flagship model for expressive audio and visuals,” highlighting three core upgrades:

• Character performance - Facial expressions, emotions, physical motion, and speech are synchronized natively rather than composited in post-processing.
• Dynamic camera motion in high-speed scenes - In chases, fight scenes, or sprints, the virtual camera tracks action dynamically while cuts maintain scene continuity.
• VFX and particle physics - Explosions, smoke, flying debris, and pyrotechnics blend seamlessly into the scene geometry rather than appearing as flat overlays.

Two distinct architecture choices set it apart from most competing tools. First, it lacks a pure text-to-video mode: generations must begin with a starting frame (Image to Video) or a set of reference assets (Reference to Video). As a result, it does not appear on text-to-video leaderboards. Second, in reference mode, it accepts up to 15 reference images and up to 3 voice audio clips to clone voice timbre, allowing consistent characters to speak with an identical voice across different sequential cuts.

Artificial Analysis leaderboard standings

Artificial Analysis ranks video models using blind head-to-head comparisons: human evaluators view two anonymous clips generated from identical prompts and choose the superior output to calculate an Elo rating. Vidu Q4 Preview debuted on October 8 at #3 on the Image-to-Video leaderboard (AA-Video-I2V v1.0). The margins at the top remain tight within standard statistical error: the 95% confidence interval for Q4 spans 1169-1189, while MiniMax H3 sits at 1173-1189. Effectively, the top six models sit in a statistical dead heat, with a clear separation only appearing from 7th place downward.

Vidu Q4 Preview debuts at #3 on the Artificial Analysis AA-Video-I2V v1.0 leaderboard, a leap from #19 for Vidu Q3 Pro

- Artificial Analysis (@ArtificialAnlys) October 8, 2026

Leaderboard figures courtesy of Artificial Analysis as of October 8, 2026. Costs per minute reflect default baseline configurations from the benchmark index.

The chart scale begins at 1000 Elo to highlight relative differences. The jump from Q3 Pro represents a 123-point Elo gain and a 16-rank ascent. Because Q4 lacks dedicated video editing and text-to-video modalities, it does not rank in those categories.

Vidu Q4 Preview specifications

Data verified via the Vidu API documentation and the official product portal. API outputs remain hosted for download for 24 hours.

How much does Vidu Q4 cost?

Vidu charges on a credit-per-second basis scaling with resolution. Generations with sound cost the exact same as silent ones, and 1080p pricing matches the previous Vidu Q3 Pro model. The baseline credit rate is $0.005 per credit.

Sources: Vidu API pricing, fal.ai (promotional discount through November 30), and WaveSpeedAI (standard billing, e.g., $0.475 for a 5-second 720p clip). The headline marketing claim of “starting at $0.014 per second” applies exclusively to lowest-tier resolutions combined with initial volume discounts, not standard 1080p rendering.

Vidu.com web subscriptions

If you generate via the web studio instead of the API, Vidu uses recurring dollar-denominated tiers. Web subscription credits are tracked independently from developer API balances. Current pricing as of October 8, 2026:

Annual rates reflect introductory sign-up discounts of 17-18%. The promotional rate of 2 credits per second at 540p is labeled as a limited-time perk; web credit burn rates for 720p and higher resolutions are only displayed inside an active account session. Paid tiers include commercial usage rights and watermark-free exports. Source: vidu.com/pricing.

Minute cost comparison: Vidu Q4 vs. rivals

To evaluate commercial viability, we look at the standardized per-minute metric provided by Artificial Analysis alongside the actual generation costs we paid for 5-second test clips in our AI Video Arena testing environment (recorded October 8 via OpenRouter across varying resolutions).

Takeaway: Vidu Q4 is not the cheapest option among top-tier models-MiniMax H3 Max holds a higher Elo score while cutting the per-minute bill by a third. However, at 1080p with integrated audio, Vidu Q4 runs noticeably cheaper than Seedance, Wan 3.0, Kling, and Veo. It also adds native 4K support, which most competitors still lack. For a deeper breakdown of the top tier, see our comparison guide: Vidu Q4 vs Seedance 2.0 vs MiniMax H3.

How to get started with Vidu Q4 step-by-step

In the web browser (no code)

1. Go to vidu.com/vidu-q4 and sign in. The web application is accessible globally.
2. Select your creation workflow: Image to Video (a single source image used as the opening frame) or Reference to Video (multiple shots of a character, product, or background environment, plus optional voice samples).
3. Upload your media and enter a prompt detailing scene movement, camera directions, spoken dialogue in quotes, and ambient sounds.
4. Choose your clip duration (3-16 seconds) and resolution. On the Free tier, 540p consumes the fewest credits.
5. Click “Create” and download the completed clip from your generation library. Note that outputs on the Free plan include a watermark and exclude commercial rights.

Via API

• Vidu API: Send a POST https://api.vidu.com/ent/v2/img2video payload with "model": "viduq4-preview" and an Authorization: Token <key> header; query task completion via /ent/v2/tasks/{id}/creations. Credit packs are purchased via platform.vidu.com.
• fal.ai: Endpoints are available under fal-ai/vidu/q4/image-to-video and fal-ai/vidu/q4/reference-to-video (supporting up to 12 reference images and 3 audio samples), billed against your card balance with a 30% discount running through November 30.
• ComfyUI: Integrated as of October 8 via ComfyUI Partner Nodes, deducted directly from your Comfy account credits.

For an end-to-end tutorial featuring full code samples, production workflows, and prompt structures, see our guide on how to generate video from images using Vidu Q4.

Benchmark tests and live examples

Transparency note: as of October 8, our team has not yet completed our own end-to-end benchmark renders for Vidu Q4. When setting up our standard test harness:

• fal.ai brought Vidu Q4 online on October 7, but our test balance was temporarily depleted.
• OpenRouter, which powers our automated AI Video Arena, does not host any Vidu models (among 30 video models available as of October 8).
• reAPI supports Vidu Q3 Pro and Q3 Turbo, but returns “Model is not supported” when calling Q4.
• The direct Vidu API requires dedicated onboarding and credit purchases.

Rather than presenting promotional footage as independent testing, we embed the manufacturer’s demo reels above and point readers to the blind comparison clips Artificial Analysis shared in their breakdown on X, which showcase Vidu Q4 head-to-head against competing models from the same start frames.

Our standard five-scene evaluation suite is ready to run:

Estimated total cost for the run via fal.ai is approximately $2.40. We will publish final benchmark scores here and in our AI Video Arena index as soon as rendering completes.

Multilingual support and non-English speech

Vidu does not publish an official directory of supported voice languages. Prompts are best written in English, with dialogue lines in target languages placed inside quotation marks. Non-English speech remains a persistent hurdle for generative video systems: when we tested foreign speech across our video benchmarks, only MiniMax H3 Max, Gemini Omni Flash 1.1, and Grok Imagine 1.5 achieved zero phonetic errors. The ability in Vidu Q4 to provide up to 3 audio reference clips offers a compelling workaround: supplying a clean native voice sample allows the model to clone timbre and prosody directly.

Who should consider Vidu Q4?

• AI filmmakers and episodic creators who need visual and auditory consistency for characters across continuous scenes-15 visual references and 3 audio samples provide wider parameter memory than most alternatives.
• Production teams targeting 4K workflows for client delivery or large-format digital signage-native 4K remains rare among top-five models.
• Action and complex VFX sequences-the development team focused heavily on dynamic cuts and particle interaction.
• Creators looking for pure text-to-video generation should look elsewhere. Without an initial reference frame, Q4 cannot generate video; look to Wan 3.0 or MiniMax H3 instead.
• Ultra-low-budget bulk production pipelines-MiniMax H3 Max yields higher benchmark performance at a lower per-minute rate, and models like Wan 3.0 allow sub-$0.25 generations at 480p.

For a broader comparison across available tools, read our breakdown of the best AI video generators and our updated AI video leaderboard.

Frequently asked questions

Is Vidu Q4 free to use?

Vidu offers a Free account tier on vidu.com that includes introductory bonus credits upon signup. Generating Vidu Q4 at 540p costs 4 credits per second on this plan. Clips produced on the free tier carry an embedded watermark and exclude commercial licensing.

How much does one minute of Vidu Q4 video cost?

Via the API, rendering 1080p with audio costs $7.20 per minute (24 credits per second at $0.005 per credit). 720p costs $5.70 per minute, while 4K runs $23.40 per minute. A 30% discount is available on fal.ai through November 30.

Can Vidu Q4 generate video directly from text prompts?

No. Vidu Q4 Preview strictly supports Image-to-Video and Reference-to-Video workflows. It requires at least one starting image frame or reference image.

What are the primary differences between Vidu Q4 and Vidu Q3?

Q4 ranks #3 on the Image-to-Video leaderboard, up from #19 for Q3 Pro (+123 Elo points). Q4 introduces 2K and 4K output options with 10-bit color alongside multi-reference support (up to 15 images and 3 audio samples). Standard 1080p pricing remains identical at 24 credits per second.

How does Vidu Q4 compare to Seedance and MiniMax H3?

On the Artificial Analysis benchmark, Vidu Q4 sits directly between MiniMax H3 and Seedance 2.0, with the score differences falling within the statistical margin of error. For an in-depth breakdown of features, video quality, and pricing differences, check our comparison: Vidu Q4 vs Seedance 2.0 vs MiniMax H3.

Is Vidu Q4 available globally?

Yes. The vidu.com website, subscription tiers, and developer API platform are accessible globally, with billing handled in US dollars. Vidu notes that features and specific tier availability may vary slightly by region.

Which third-party platforms host Vidu Q4 via API?

Vidu Q4 is currently accessible through fal.ai, WaveSpeedAI, and ComfyUI Partner Nodes. It is not currently hosted on OpenRouter, Replicate, or reAPI (as of October 8).

When will the full version of Vidu Q4 launch?

ShengShu Technology has not announced an official date for the general release. The Preview build is intended to gather feedback from active production pipelines before locking final release specifications.

Related articles and resources

• Vidu Q4 Preview lands at #3 on the global AI video leaderboard
• Vidu Q4 vs Seedance 2.0 vs MiniMax H3: Which model should you choose?
• How to generate video from images using Vidu Q4: Practical guide
• AI Video Arena: Comparing leading video models across standardized prompts