On October 7, 2026, Shengshu Technology launched Vidu Q4 Preview, the latest flagship from the Tsinghua-founded lab behind the Vidu model family. Its pitch is different from most video models: not just prettier motion, but performance. Upload a photo, and Q4 aims to turn it into a clip that feels acted — coordinated facial expression, emotion, body movement, and voice — rather than animated.
Within 24 hours it ranked third on Artificial Analysis's image-to-video leaderboard. Here's what matters if you build with video models.
Built to Act, Not Just Animate
Q4 coordinates expression, emotion, movement, and voice in one generation. Effects like explosions and particles sit inside the scene instead of floating over it, camera moves hold together through fights and chases, and a character's voice stays consistent with natural emotional delivery.
The output specs: 540p up to 4K with 10-bit color at the top tiers (over a billion colors versus 16.7 million in 8-bit — smoother gradients on skin, sky, and lighting), clips of 3 to 16 seconds at 24 fps, keeping the aspect ratio of your source image, with audio on by default.
The Reference System: Consistency as Input
Like keyframes and reference systems elsewhere in the October wave, Q4 treats consistency as something you supply, not something you hope for:
- Up to 15 reference images — build complex scenes and keep characters, outfits, and settings consistent across shots.
- 3 reference audio clips — pin a character's voice, with stronger dialogue delivery and lip-sync.
- Natural-language camera direction — describe push, pan, tilt, and multi-shot sequences in plain words and let the model plan the motion.
For episodic content — AI short drama is Q4's headline use case — consistent characters across shots is the feature that decides whether a workflow is viable at all.
The Leaderboard Debut
On October 8, 2026, Vidu Q4 Preview ranked third on Artificial Analysis's image-to-video leaderboard at Elo 1,179 — up 123 points from Vidu Q3 Pro's 1,056, within a day of launch. That's a top-tier debut by any measure, and it positions Q4's edge as reference-driven acting and multi-image consistency rather than pure prompt-to-video generation.
The market noticed fast: Perfect Corp integrated Q4 Preview across the YouCam apps (1.1 billion+ downloads) and its online editor within days, and platforms like SelfyzAI added it to their model pickers in the first wave.
Pricing: Per-Second, With a Launch Deal
- Launch promo: from $0.014 per second of video
- List prices: $0.045/sec at 540p, $0.12/sec at 1080p, up to $0.39/sec at 4K
- 30% off image-to-video and reference-to-video through November 30, 2026, plus half-price off-peak jobs
At list rates, a 10-second 1080p clip runs about $1.20 — mid-pack for the industry, with launch pricing undercutting most rivals. You can try it at platform.vidu.com.
The Catches
Three honest limits. First, this is a preview: behavior can change, and independent long-term testing is still thin. Second, there is no text-to-video mode yet — Q4 Preview is image-to-video and reference-to-video only, so every generation starts from a photo or references. Third, budget against list prices for anything beyond the trial window; the $0.014/sec rate is promotional.
Who should try it: short-drama and short-form creators who need the same character in every shot, advertisers turning product photos into promos, music-video creators who live or die on lip-sync, and filmmakers using plain-English camera direction for fast previz.
This is a condensed explainer. The full breakdown — feature-by-feature analysis, the complete pricing table, head-to-head comparison with Veo, Runway, and Kling, and FAQ — is on NeurAITool: Vidu Q4: The AI Video Model That Turns Photos Into 4K Cinema. For the wider landscape, see the 90+ best AI video generator tools directory.
