A AI Tool Testing

Review · ai creative

Runway Gen-4.5 Review: The Control Platform, If You Can Afford the Credits

Runway Gen-4.5 gives filmmakers the best camera control in AI video. It no longer tops the benchmarks, output is silent 720p, and the credit math is brutal. Full synthesis.

By Max Langley ·

Disclosure: We earn commissions from links on this page, as an Amazon Associate we earn from qualifying purchases, at no extra cost to you. This never affects what we recommend. Read our editorial standards →

Runway

Runway Gen-4.5

8.0/10

Synthesis score

Standard $12/mo; Pro $28/mo; Max $76/mo (annual billing)

The control platform: if a shot needs a specified camera move and a consistent character, nothing else obeys like Gen-4.5. But output is silent 720p capped at 10 seconds, the benchmark crown passed to newer models, and $28 a month buys about 90 seconds. Professionals only.

Pros

  • +The deepest directorial control in the category: natural-language camera moves (tracking, dolly, crane, rack focus, handheld shake), motion brush, and reference-driven character consistency across cuts
  • +Debuted at #1 on the Artificial Analysis text-to-video leaderboard (Elo 1,247 at its December 2025 launch, then ahead of Veo 3 and Sora 2 Pro), per Runway's own dated research post
  • +Gen-4.5 fixed the 'floaty physics' of earlier versions; objects now carry realistic weight and momentum per AdCreate's testing
  • +Paid tiers bundle the wider toolchain (Act-Two performance capture, the Aleph video editor) plus third-party models including Veo 3.1, Kling 3.0 Pro, and Seedance 2.0, so one subscription hedges across the big models

Cons

  • The credit math is the story: Gen-4.5 burns 25 credits per second, so the $12 Standard plan yields about 25 seconds a month, Pro ($28) about 90 seconds, and Max ($76) about 380 seconds
  • Output is silent, 720p, and capped at 10 seconds per generation (2 to 10 seconds, per Runway's own docs): no native audio, the shortest ceiling of the big three after Veo, and no 4K render
  • No longer the benchmark leader: by mid-2026, ByteDance's Seedance 2.0 and Alibaba's HappyHorse models had displaced Gen-4.5 from the top of the Artificial Analysis board
  • Failed or defective generations still consume credits, and regenerating a flawed clip charges you twice; Standard-plan users are also locked out of custom voices for lip-sync
  • Text rendered inside videos remains unreliable, hands still artifact, and per Magic Hour's analysis Runway runs roughly 40% more expensive than Kling per second at comparable quality

Disclosure: This review is a synthesis of independent testing coverage, pricing analyses, and vendor documentation. We have not independently generated video with Runway. We have no paid relationship with Runway, and our link is the plain vendor URL.

Runway Gen-4.5 is the model that launched with the benchmark crown and lives by the credit meter. It debuted at #1 on the Artificial Analysis text-to-video leaderboard in December 2025; by mid-2026 newer models from ByteDance and Alibaba had taken the top of that board. In our best AI video generator guide it holds the professional-control slot rather than the best-overall slot, and the case for it was never really the leaderboard anyway.

Gen-4.5 shipped in December 2025, and its pitch is different from every rival’s. Kling sells motion realism, Veo sells accessibility; Runway sells obedience. It is the platform for people who know what shot they want before they generate it.

What obedience means in practice

Three capabilities define it. First, camera grammar: Gen-4.5 understands tracking shots, dolly moves, crane shots, handheld shake, and rack focus specified in plain language. If a storyboard says “slow dolly-in on the subject, background falls out of focus,” this is the only model that treats that as an instruction rather than a suggestion. Second, the motion brush lets you direct movement within a frame, painting which elements move and how. Third, reference-driven character consistency holds a face, wardrobe, and build steady across generated cuts, which is the difference between a sequence and a slideshow of similar-looking strangers.

Know the output envelope before buying, though, because it is narrower than the control story suggests. Per Runway’s own documentation, Gen-4.5 generates 2 to 10 seconds per pass at 720p, and the output is silent: no native audio of any kind. That makes it the only one of the big three without in-model sound (Veo 3.1 and Kling’s Omni variant both generate audio), and its 10-second ceiling now trails Kling’s 15-second window. What survives contact with the spec sheet is the physics and the obedience: AdCreate’s testing confirms the floaty-physics complaint from earlier generations is substantially fixed, with weight and momentum rendered credibly. Plan the missing pieces as separate line items: working creators typically add narration with ElevenLabs and cut the final piece in Descript (see our ElevenLabs review and Descript review).

The benchmark history deserves precise tenses. At launch in December 2025, Gen-4.5 held #1 on Artificial Analysis at Elo 1,247, ahead of Google Veo 3 (1,226) and OpenAI’s Sora 2 Pro (1,206), a claim Runway’s own research post dates to November 30, 2025. By mid-2026 it had been displaced by ByteDance’s Seedance 2.0 and Alibaba’s HappyHorse and Wan models and no longer appears near the top of that board. Our guide is deliberately skeptical of leaderboard-first buying, and this is why: crowns in this category change heads in months. The control toolset is the durable asset.

The credit math, stated plainly

Here is the entire catch, and it is not small. Gen-4.5 burns 25 credits per second of output. Runway’s plans, billed annually: Standard at $12 a month includes 625 credits, about 25 seconds of Gen-4.5 a month. Pro at $28 includes 2,250 credits, about 90 seconds. Max at $76 includes 9,500 credits, about 380 seconds. Per-second cost falls from roughly $0.48 on Standard to $0.20 on Max.

Twenty-five seconds a month barely covers one polished clip. Even Pro’s 90 seconds amounts to six to nine short clips, a weekend of material for an active creator, before counting iteration. And iteration counts double here: failed or defective generations consume credits at full rate, so a clip that takes three attempts costs three clips. Standard users additionally lose custom voices for lip-sync and text-to-speech. Magic Hour’s cross-platform analysis prices Runway roughly 40% above Kling per second at comparable quality.

One genuine mitigation: paid tiers now bundle third-party models, including Veo 3.1, Kling 3.0 Pro, and Seedance 2.0, plus image models like FLUX.2. Routine shots can burn cheaper credits on an alternate model, saving Gen-4.5 for the hero frame. One subscription becomes a hedged portfolio of the field’s best models, which is a quietly strong reason to hold a Runway plan even if Gen-4.5 is not your volume workhorse.

Remaining rough edges

Text inside generated video is still unreliable; assume signage and labels need compositing. Hands and fine motor actions still artifact under scrutiny. And precise product renders want image references rather than text prompts. None of these are unique to Runway, but a platform selling control gets judged by control standards.

Against Kling and Veo

Against Kling 3.0: Runway wins on control, full stop. Kling wins on price, volume, the most natural organic motion (water, fabric, crowds especially), and now duration and audio too: 15-second generations with native sound against Runway’s 10 silent seconds. Less than half the working cost buys most of the quality, which is why Kling takes our guide’s best-overall slot while Runway takes the professional one.

Against Google Veo 3.1: Veo’s free tier and native audio make it the on-ramp, and on sound Veo simply does something Runway does not. Runway’s answer is direction: Veo’s 8-second cap, character drift, and hands-off camera keep it out of the production conversations Gen-4.5 was built for. Notably, Runway’s own platform includes Veo 3.1, so this comparison can be run inside one subscription.

Verdict

If your work specifies shots, Gen-4.5 has no real substitute in mid-2026: the camera control and character consistency are the professional toolkit, and they survive the loss of the benchmark crown. Buy it knowing the envelope: 2 to 10 silent seconds at 720p per generation, with sound and length handled elsewhere in your pipeline. Price it as a production line item rather than a personal subscription: the $76 Max tier is where the credit math starts resembling a workflow, and professionals billing clients will clear that bar without blinking. Everyone else should draft on Kling or Veo and reach for Runway when the shot must be right.

Frequently asked questions

How far do the credits actually go?
Gen-4.5 costs 25 credits per second of output. Standard ($12/month, 625 credits) is about 25 seconds; Pro ($28, 2,250 credits) about 90 seconds; Max ($76, 9,500 credits) about 380 seconds. Those are ceilings before iteration: failed generations charge full credits, so a working assumption of half your nominal seconds is realistic. Cost per second falls from $0.48 on Standard to $0.20 on Max.
Who is Runway actually for?
People who direct shots rather than generate clips: ad and film production workflows where a specific lens move, a consistent character, and continuity across cuts are requirements, not preferences. If 'generate and pick the best of six' describes your workflow, Kling delivers more volume per dollar. If a client brief specifies the shot, Runway is the only model that reliably obeys.
What is Gen-4.5 measurably best at?
Control, and it needs saying precisely because the benchmark story has aged. Gen-4.5 launched at #1 on Artificial Analysis (Elo 1,247, December 2025) but had been displaced by Seedance 2.0 and Alibaba's HappyHorse models by mid-2026. Its durable edge is camera grammar specified in natural language, motion brush for directing movement within a frame, and reference-driven character consistency. Its physics also improved sharply over Gen-4's floaty look, per AdCreate's testing.
Does a Runway subscription include other companies' models?
Yes, and it is an underrated reason to pick it. Paid tiers include access to third-party models, including Google Veo 3.1, Kling 3.0 Pro, and Seedance 2.0, alongside image models like FLUX.2. You can burn cheaper credits on an alternate model when a shot does not need Gen-4.5, which partially softens the credit constraint.
Runway or Kling?
Runway for control; Kling for value, volume, duration, and audio. Kling generates up to 15 seconds per shot with native audio from $6.99 a month intro pricing; Runway generates 2 to 10 silent seconds at 720p, with far better obedience to camera and character direction. Our guide's framing: professionals who storyboard pick Runway, iterators pick Kling. Many working creators run Kling for drafts and Runway for the hero shot.

Sources

Every claim in this guide that isn't first-person experience is traceable to one of the sources below. URLs verified at publication; some may rot. Let us know if so.

  1. Runway Gen-4.5 Review: Features, Pricing (2026)AdCreate, 2026Testing source for camera-control grammar, physics improvements over Gen-4, character consistency, text-rendering and hand-artifact weaknesses, launch-time Elo figures, December 2025 launch. Its one-minute-duration, 4K, and native-audio claims are contradicted by Runway's own documentation and are not used here.
  2. Creating with Gen-4.5Runway Help Center, fetched July 10, 2026Primary source for the output envelope: supported durations 2–10 seconds, 720p output resolution, supported aspect ratios. No audio capability documented.
  3. Introducing Runway Gen-4.5Runway Research, November 30, 2025Runway's own launch post: 1,247 Elo and #1 Artificial Analysis placement, explicitly dated November 30, 2025. No mention of audio generation.
  4. Text to Video LeaderboardArtificial Analysis, fetched July 10, 2026Current standings: Seedance 2.0 #1 (Elo 1,224), Alibaba Wan and HappyHorse models following, Kling 3.0 Pro at 1,109, Veo 3.1 at 1,095; Gen-4.5 absent from the with-audio board.
  5. Runway AI pricing in 2026: every plan, credit cost, and what you actually geteesel AI, fetched July 10, 2026Plan prices and credit allotments, 25 credits/second Gen-4.5 burn rate, seconds-per-tier math, failed-generation charging, Standard-tier custom-voice lockout, bundled third-party models.
  6. Runway Gen-4.5 Pricing: Plans, Credits & Free Trial Explained (2026)Magic Hour, 2026Cross-check on tier pricing and the roughly 40% per-second premium versus Kling at comparable quality.
  7. Runway PricingRunway, fetched July 10, 2026Vendor source of record for current plan prices and included models. No affiliate relationship; our link is the plain vendor URL.