Vidu Q4 Preview Ranks No. 3 in Image-to-Video, Lacks Text Mode

A big jump on a human-judged leaderboard

Shengshu Technology, the Beijing-based generative AI company spun out of Tsinghua University's AI labs, has released Vidu Q4 Preview, an early public build of its next flagship video model.5 Its most notable result came from an independent benchmark. The model entered Artificial Analysis's AA-Video-I2V v1.0 image-to-video leaderboard at No. 3 with an Elo score of 1,179.1 Shengshu's previous best performer, Vidu Q3 Pro, sits at No. 19 with 1,056, so the new model climbed 16 places and gained 123 points.1 Artificial Analysis builds these rankings from pairwise human preference votes between generated clips.1

The launch was announced on October 7, 2026, through releases datelined in both Singapore and Beijing.45 Chinese tech outlet Pandaily covered it a few days later.2

What the model does

Across every account, Shengshu's main pitch is acting. The model is designed to coordinate facial expression, emotion, body movement and voice so performances look less mechanical.4 It generates picture and sound in a single pass, which means lip movement and line delivery are meant to match without separate audio work.3 Shengshu also says it has improved dynamic camera movement, shot changes and transitions between complex shots.2

The specifications largely agree from source to source. Clips run 3 to 16 seconds.13 Resolution options are 540p, 720p, 1080p, 2K and 4K, with 10-bit color depth.25 Native synchronized audio is included.1 To keep characters consistent, Pandaily and AlphaSignal both report support for up to 15 reference images, plus up to three audio references to hold a character's voice steady.12 One third-party studio page lists a ceiling of 12 image references instead.3 That may simply reflect how a particular integration exposes the feature.

Shengshu is aiming at professionals who produce video in volume. It names short-drama production companies, advertising agencies, social content creators and film teams as its core audience.5

The missing piece: text-to-video

The main limitation is the input mode. For now, Q4 Preview supports only image-to-video and reference-to-video generation.12 Users cannot type a prompt and get a scene from scratch. A text prompt can guide a generation, but only alongside a source image or references.3

That context matters for the leaderboard result. The No. 3 placement is specific to the image-to-video category. It says nothing yet about how Vidu Q4 would compare in text-to-video, which is arguably the more visible battleground. Leaning into image and reference workflows makes strategic sense for Shengshu's stated customers. Short-drama and e-commerce teams usually begin with existing character art or product shots and need many consistent takes of them. Even so, a flagship model without text-to-video is incomplete, and the "Preview" label suggests more is coming.

Pricing: cheaper, but the figures need care

Price is Shengshu's other major selling point, and the numbers vary depending on the channel. The press release advertises launch pricing starting at $0.014 per second.4 Pandaily cites a two-month launch offer on Vidu's own SaaS product starting at RMB 0.09 per second.2 AlphaSignal gives a standard 1080p rate of $7.20 per minute and notes a 30% discount through November on web and API access.1

The framing of the comparison with Q3 Pro is also inconsistent. AlphaSignal's headline says Q4 costs 25 percent less per minute than Q3 Pro. Its summary, however, describes the 1080p price as matching Q3 Pro while delivering better quality.1 Shengshu's own claim, repeated on a partner page, is up to five times as much output for the same budget under comparable settings.3 CEO Yihang Luo described the strategy in terms of iteration, saying efficiency gains should give creators the freedom to try one more shot or one more version.4

The most cautious reading is this. Headline per-second rates are promotional floors, probably at lower resolutions and during the launch window. The steady-state price for high-resolution output looks roughly in line with, or modestly below, the previous generation.

The takeaway

Vidu Q4 Preview is a real step up for Shengshu. The independent human-preference ranking carries more weight than vendor claims, and a 16-place jump is hard to dismiss. Its focus on consistent characters, synced voice and cheaper retakes fits the high-volume short-video and advertising market it is chasing. Still, this is a preview with a narrow scope. Until text-to-video arrives and the promotional pricing ends, buyers should treat the No. 3 ranking as strong evidence for one workflow, not as a verdict on the whole model.