Kling 4.0 vs Vidu Q4 Preview: One of Them Cannot Take a Text Prompt

Open Vidu Q4 Preview's product page and count the generation modes. There are two: Image-to-Video and Reference-to-Video. There is no text-to-video tab, and the API documentation matches — the endpoints are /ent/v2/img2video and /ent/v2/reference2video, and neither accepts a prompt alone. Kling 4.0, by contrast, is a text-first model that also takes images, first and last frames, and up to ten keyframes. Before any comparison of Elo, resolution or price, that is the fact that decides most projects: one of these models can start from a sentence, and the other needs a picture.

Everything else in this matchup is downstream of that. Vidu Q4 Preview is a targeted instrument for shots that begin with an image. Kling 4.0 is a general-purpose generator with a director's control surface. They overlap on image-to-video and diverge sharply everywhere else.

What Shengshu shipped on 7 October

Vidu Q4 Preview is the first public preview of Vidu's next-generation flagship, released deliberately ahead of the full Q4 model so that feedback can shape the final release. It generates video with native audio — on by default, dialogue and sound effects included — and its specification is unusually generous at the reference end.

• Modes — Image-to-Video, 3 to 16 seconds; Reference-to-Video, 1 to 16 seconds. No text-to-video.

• Resolution — a ladder of 540p, 720p, 1080p, 2K and 4K, with 2K and 4K carrying 10-bit colour depth.

• References — up to 15 reference images and up to 3 reference audio clips of three to twelve seconds each.

• Prompt length — up to 20,000 characters on the image-to-video endpoint.

• Audio — native, defaulting to true. You switch it off deliberately rather than on.

• Return URLs — creation URLs stay valid for 24 hours, which matters if you batch and render later.

The vendor's efficiency claim — "up to five times as much output for the same budget" under comparable specifications — is a marketing construction until somebody reproduces it, and Shengshu's own launch material carries a caveat that final pricing, supported resolutions, feature availability and usage terms may vary by plan and region.

What Kling 4.0 offers on the same axis

Kling 4.0 has been available to everyone since 15 October 2026. On the image-to-video axis it is the more flexible of the two: up to 10 keyframe images you can place to define character states and scene changes, a 15-item reference budget split across 10 images, 5 videos capped at 30 seconds combined, voice references and 7 elements, and a single-pass duration of 3 to 30 seconds.

The duration gap is the one that shows up in the edit. Vidu Q4 Preview stops at 16 seconds. Kling 4.0 runs to 30 in one pass, with extension toward 2 minutes listed as coming soon. For a 20-second shot, Vidu needs a join and Kling does not. For a 10-second shot — the actual unit of most social and ad work — the difference disappears.

Resolution is closer than the marketing suggests. Both claim 4K. Vidu's 2K and 4K tiers ship with 10-bit colour depth today; Kling's 10-bit HDR at 1080p and 4K is marked coming soon on Kuaishou's own release note and feature page, alongside native 4K with 21:9. If a ten-bit deliverable is a hard requirement this quarter, Vidu's claim is the one that is currently shipping, subject to the plan and region caveat.

The boards say something narrower than the launch copy

Artificial Analysis runs video leaderboards with separate Elo scales and separate vote pools, and they are versioned. At its October debut on the v1.0 image-to-video board — image-to-video, audio kept in — Vidu Q4 Preview entered third at Elo 1,179 ±10 over 5,543 samples, at a tracked $7.20 per minute. Above it sat MiniMax H3 Max at 1,195 and MiniMax H3 at 1,181; Gemini Omni Flash was fourth at 1,178. The ten-point intervals from third through sixth overlapped almost completely, so "third" was a position inside noise even then.

The gap that was not inside noise was the one to Vidu's predecessor. On that capture, Vidu Q3 Pro sat nineteenth at 1,056. A 123-point move in one release is larger than the entire spread across the board's top six, and it is the strongest evidence that this preview is a real generational step rather than a rebrand.

Then the board moved. The version I read on 23 October is AA-Video-I2V v1.1, and it lists 24 models: MiniMax H3 first at 1,140, FLUX 3 Max second at 1,138, Dreamina Seedance 2.5 third at 1,138, MiniMax H3 Max sixth at 1,125, and Vidu Q3 Pro seventeenth at 1,016. Vidu Q4 Preview does not appear anywhere in that list. The note at the top of the board says three models were added in the last 30 days, and it names none of them as Vidu's preview.

I am not going to pretend that means the model got worse, and I cannot tell you why it is missing — whether it was withdrawn, reclassified, or simply not carried into the new board version. What it does mean is that every "third in the world" claim about this model, including the one this article opened with, is a statement about a specific capture on a specific board version, and that the board is not currently corroborating it. Treat the rank as history and the product page as the durable fact.

On the text-to-video board, Vidu Q4 Preview has never appeared — and that absence is not a scoring problem. The best Vidu entry there is Vidu Q3 Pro, mid-table, while Wan 3.0 leads. It is consistent with the product page: this is an image-and-reference model, and the text-to-video board is not measuring it.

Kling 4.0 appears on none of these boards either, eight days after general availability. That is not a low score; it is no score, in either direction. So the comparison of quality between these two models does not exist in any shared unit, and the only independent number in this article belongs to a model that the current leaderboard no longer lists.

The two price numbers, and the one you should budget from

Shengshu advertises Vidu Q4 Preview "from $0.014 per second," which is $0.84 per minute. The October Artificial Analysis capture recorded the same model at $7.20 per minute. Those are not the same product: the launch figure is a promotional floor, almost certainly at the bottom of the resolution ladder, and the tracked figure was a measured rate at a production resolution. The vendor also says outright that pricing is promotional and may vary by plan and region. Budget from the higher number and treat the lower one as an acquisition offer.

Kling 4.0's published preview rates are $0.157 per second at 720p and $0.235 at 1080p — $9.42 and $14.10 per minute — with no 4K rate at all and a page that still marks the model Coming Soon. On those numbers the two models are in the same band at production resolution, with Vidu cheaper and Kling buying you 14 extra seconds per generation.

• Per 10-second 1080p clip — Vidu at the tracked $7.20/min is about $1.20. Kling 4.0 at $0.235/s is $2.35.

• Per 30 seconds of finished footage — Vidu needs two generations and a join, roughly $3.60 plus edit time. Kling 4.0 needs one, at $7.05.

• Per 100 ten-second clips — Vidu about $120; Kling about $235; the base MiniMax H3 route mentioned below is about $80.

Access: one is callable, the other is mid-rollout

Neither model is a clean self-serve default. Vidu Q4 Preview has a real API with documented endpoints, which makes it more buildable today than Kling 4.0, whose API began opening in waves from 21 October while its developer price sheet still says the model is not yet available. But Vidu's build is explicitly a preview: promotional pricing, feature availability subject to change, and even a mistyped model alias — viduq4-previerw — alongside the correct one in the documentation. That is a small thing that tells you something true about where this build sits.

For a team that needs both a text-first and an image-first path, the practical shape is to keep the choice at the routing layer rather than in the application code. OrcaRouter does not host either model — there is no Vidu identifier in the catalogue and no Kling 4.0 route — but it does host Kling 3.0 at $0.084 per request and MiniMax H3 at $0.08 per second of generated video, on one OpenAI-compatible endpoint across more than 200 models at provider list price with 0% markup. The routing DSL is the part that fits this particular matchup: it lets you put a text-first model and an image-first model behind one interface so the decision becomes a parameter, and automatic failover means a preview route going down does not take the batch with it.

Choosing between them is mostly a question about your first frame

If your shot begins with a sentence, Vidu Q4 Preview is the wrong tool and no amount of Elo changes that. You would be generating a still elsewhere and then handing it over — which is a legitimate pipeline, and one that many teams run deliberately, because it gives them control of the opening frame.

If your shot begins with a photograph, a product image, a character sheet or a storyboard frame, Vidu Q4 Preview is the better-evidenced choice of the two: it is third on the board, it improved 123 Elo over its own predecessor, it takes 15 reference images and 3 audio clips, and its 10-bit 4K is shipping rather than promised. The 16-second ceiling is the price.

If you need length, keyframe choreography, or a text-to-video path with a reference budget, Kling 4.0 is the only one of the two that offers them — and you should treat its price as provisional until Kuaishou publishes a 4K rate and clears the Coming Soon banner.

What would change this picture is not another spec. It is a Kling 4.0 entry on the image-to-video board — the one place these two models could actually be compared in the same unit — and a full Vidu Q4 release that either keeps the 15-image reference budget and the 10-bit 4K or quietly narrows them when the promotional pricing ends. Until both exist, the honest ranking is the one the boards support on the day you read them: Vidu Q4 Preview was a measured third in October and is not currently listed, Kling 4.0 is unmeasured everywhere, and your first frame decides which question you are even asking.