Vidu Q4 Preview vs Runway Gen-4.5: One Is Third on a Public Board, the Other Is Not On It

There is a version of this comparison that tallies two rate cards and calls the cheaper one the winner. That version is available everywhere and it is not very useful, because Vidu Q4 Preview and Runway Gen-4.5 are priced in units that only look comparable. Runway Gen-4.5 bills 12 credits per generated second, with a credit worth a cent on the developer tier — about $0.12 per second of output, and a documented surcharge of five more credits per second for ProRes or PNG sequence export. Vidu Q4 Preview, launched 7 October 2026 by Shengshu Technology, bills per second by resolution tier, from 9 credits per second at 540p to 78 at 4K on Shengshu's own platform.

The more useful comparison starts somewhere else: on the image-to-video board at Artificial Analysis, read 8 October 2026, Vidu Q4 Preview sits third at 1,179 Elo over 5,543 votes. Runway does not appear in the top twenty-eight rows of that board. That is not a claim that Gen-4.5 is bad — it is a statement about what each vendor has chosen to publish and submit, and it changes what kind of decision this actually is.

The evidence is not symmetric, and that matters more than the specs

A preference board is not a verdict, but a model's absence from one is information. Vidu Q4 Preview has accumulated 5,543 human pairwise votes on AA-Video-I2V v1.0 within days of its 7 October launch, sitting in a tie with the two MiniMax builds at the top. Runway Gen-4.5 has no row there.

What Runway does publish, on its own research page, is a capability claim rather than a score: Gen-4.5 is described as the company's most advanced video generation model, with improvements in motion quality, prompt adherence and temporal consistency, available on all paid plans. Its help documentation, as reported by third-party pricing trackers in September 2026, lists two-to-ten-second generations at 720p output and 24 or 25 frames per second.

So the honest framing of this matchup is a measured model against a claimed one. Every number for Vidu Q4 Preview on the board carries a sample count and a confidence interval; every number for Gen-4.5 on this page is either the vendor's own or a secondary source's arithmetic against the vendor's rate card. That asymmetry is not a knock on Runway — plenty of shipping production tools skip the arena — but a reader comparing the two should know which side of the page has been independently voted on.

The spec line, with resolution as the real fault line

• Maximum output resolution — Vidu Q4 Preview: 4K, with 540p, 720p, 1080p, 2K and 4K as generation tiers vs Runway Gen-4.5: 720p generation, with a plan-level 4K upscale entitlement that is a separate operation and does not make the original generation native 4K

• Clip length — Vidu Q4 Preview: 3–16 seconds on Image-to-Video and 1–16 seconds on Reference-to-Video vs Runway Gen-4.5: 2–10 seconds

• Frame rate — Vidu Q4 Preview: not published as a selectable parameter in the image-to-video endpoint vs Runway Gen-4.5: 24 or 25 fps

• Reference inputs — Vidu Q4 Preview: up to 15 reference images plus up to 3 reference audio clips vs Runway Gen-4.5: a single source frame for image-to-video, with video-to-video handled by Runway's separate Modify Video tooling

• Audio — Vidu Q4 Preview: generated alongside the video via an audio parameter that outputs dialogue and sound effects vs Runway Gen-4.5: native audio on the model, plus separate Runway audio tools

• Colour — Vidu Q4 Preview: 2K and 4K at 10-bit colour vs Runway Gen-4.5: HDR profiles documented on the developer side with their own surcharges

The resolution line is the one that decides real projects. If your delivery specification is 720p for social, Gen-4.5's ceiling is not a limitation. If it is 4K for a client master, Vidu Q4 Preview generates there natively and Gen-4.5 gets there through an upscale — which is a different thing, and the difference shows in fine detail rather than in the export label. Compare downloaded dimensions rather than file names.

Doing the credit arithmetic without crossing the two rate cards

Both models bill per second of output, so the comparison is arithmetically clean in a way that per-clip pricing is not. What is not clean is the unit.

Runway Gen-4.5's 12 credits per second is a single rate that does not vary with the output setting you choose, because the output setting barely varies: 720p, two to ten seconds. A five-second generation is 60 credits, $0.60 on the developer tier at $0.01 per credit. Add ProRes or PNG sequence output and it is 85 credits, $0.85. A ten-second generation is 120 credits, $1.20. Every one of those figures is predictable before you press generate, which is a real property and one worth paying something for.

Vidu Q4 Preview's rate varies by tier by nearly a factor of nine: 9 credits per second at 540p, 19 at 720p, 24 at 1080p, 38 at 2K, 78 at 4K. On Shengshu's platform the standard credit rate is published at $0.005, with limited-time package discounts of up to 30% running alongside it. At list rate that is roughly $0.045 per second at 540p rising to about $0.39 at 4K, or $0.72 for a sixteen-second 4K clip before any discount. The $0.014 per second headline from the launch announcement is the 540p tier at a discounted credit rate.

Laid out honestly, the comparison is: Runway charges a single mid-tier rate for a 720p ten-second ceiling, and Vidu charges a sliding rate for a 4K sixteen-second ceiling whose floor is far below Runway's flat rate and whose top is more than three times it. If you never leave 720p, Runway's $0.12 per second is roughly a quarter more than Vidu's 720p tier at $0.095 per second, and you get the ProRes surcharge and the 24/25fps parameter for it. If you work at 4K, there is no comparison to make, because only one of the two generates there.

One arithmetic both brands expect you to do yourself: accepted footage is not generated footage. Runway's own pricing documentation distinguishes attempted generations from accepted footage, and Vidu's launch material makes the same assumption from the other direction — CEO Yihang Luo's framing that efficiency gains exist to let creators "try one more shot or make one more version." Budget per delivered clip, including the takes you discard.

Where each one is the only answer

Take Runway Gen-4.5 when the deliverable is a shorter, lower-resolution clip inside a Runway workflow, or when you need professional output formats without a resolution fight — ProRes and PNG sequence export are documented and priced, and the surcharge is disclosed. Its 24/25fps parameter is also a genuine production requirement that Vidu Q4 Preview does not expose. And the flat per-second rate makes a 720p budget a one-line calculation regardless of the settings you choose.

Take Vidu Q4 Preview when the job involves a cast that has to hold together, a voice that has to stay the same character across a take, native 4K output, a sixteen-second window, or a cheap 540p pass to block a shot before committing to a full-resolution render. Reference-to-Video's 15-image and 3-audio-clip budget is aimed squarely at the continuity problem, and nothing in Gen-4.5's published specification competes with it on that axis.

What would change this: Runway submitting Gen-4.5 to the image-to-video board would replace this page's central asymmetry with an actual number, and a Runway release that raises the generation tier past 720p would remove Vidu's resolution advantage. On the other side, the full Vidu Q4 release is where Shengshu's own material says pricing and feature availability settle — a GA build that holds its board position at the same per-second curve makes it a production commitment rather than a preview one.

Calling video models without a second contract

OrcaRouter serves video and language models on one OpenAI-compatible endpoint behind one key, at provider list prices passed through with no markup added, so a vendor rate change shows up here the same day rather than at renewal. Neither Runway Gen-4.5 nor Vidu Q4 Preview is currently an OrcaRouter route. What we do route is Kling 3.0 in its standard, Turbo and v2.6 variants, alongside the language models, which is the arrangement that makes evaluating several video models at once cheap: one key, one request shape, and automatic failover between routes instead of a separate contract for every model you want to test.

The practical version of that for this matchup: the thing that will change your mind about Runway versus Vidu is probably your own acceptance rate at your own settings, and the way to learn it is to run both rather than to read either rate card twice.