Runway review: usable video outputs and editing

Editor's note: This is a documentation-led edition. We have not yet run our own hands-on tests of Runway. What follows rests on official documentation, published pricing and court records we read in full, all accessed on 16 October 2026. A tested update will follow.

Runway generates clips from text or a still, but much of what it now sells is about changing footage you already have: swapping a product, relighting a shot, widening a vertical clip. The useful question for a creator is not whether it can make something striking. It is whether it can make a shot you can cut into a real edit, at a cost you can plan, under terms you can accept.

This review answers that from the record: current models, what each plan buys in seconds, clip limits, editing tools, ownership and training terms, provenance, enterprise, documented limits and the artists' lawsuit naming Runway. We give no quality verdicts or scores until we test. Our shortlist of AI video generators for creator workflows places Runway among the alternatives.

What Runway is, and what each tool is for

Runway is a browser studio with a model menu, task-specific "Apps", an agent and a developer API, Runway Dev. The same credits also buy models from Google, Kling, ByteDance and others.[1][2] This review concentrates on Runway's own four video models:

• Gen-4.5 is the flagship generator, with text-to-video and image-to-video. Its help page says it "currently offers Text to Video and Image to Video control, with support for additional inputs coming soon."[3]
• Gen-4 Turbo is the cheaper image-to-video model; Runway advises exploring in Turbo first. Gen-4 is now labelled "an older generation".[4]
• Aleph 2.0 is the editing model behind Edit Studio. You change one frame, and it carries that change through the clip. Runway's pitch: "Change only what you want, keep everything else."[5][6]
• Act-Two transfers a recorded performance, including face, speech and, with a character image, gestures, onto a character.[7]

Gen-3 Alpha was retired on 8 July 2026 and Gen-3 Alpha Turbo on 30 July 2026, so older tutorials describe tools you can no longer use.[8]

At launch on 1 December 2025, Runway said Gen-4.5 ranked first on the Artificial Analysis text-to-video leaderboard with 1,247 Elo points.[9] That is a vendor-reported human-preference vote at one moment, not evidence that your shot will work. Our explainer on how AI video generators work covers how to read such comparisons.

Plans and credit costs (as of 16 October 2026)

Sources: Runway pricing, credits FAQ, free-plan and Team pages.[1][10][11][12] Paid plans can buy packs of at least 1,000 credits, which never expire. App and API credits are separate balances.[10]

Per-second rates are consistent across the help centre, Academy and API pricing:

Sources: help centre model guides, Runway Academy model table and Runway Dev pricing.[3][4][6][7][2][13]

What that buys. On our arithmetic, Standard covers about 52 seconds of Gen-4.5, 125 of Gen-4 Turbo or 22 of Aleph 2.0; Pro about 187 seconds of Gen-4.5 or 80 of Aleph 2.0. That is generated footage, not finished footage: rejected takes cost the same as kept ones. One conflict: the pricing page converts Standard and Pro into 104 and 375 seconds of Turbo, implying 6 credits a second, while the model guide and API say 5.[1][4] Check the in-app estimate.

At list prices, a Gen-4.5 second costs about $0.29 on monthly Standard, $0.19 on Pro and $0.12 on Max if every credit is used; API credits at $0.01 also give $0.12.[13] These are our calculations, not Runway's figures.

Clip limits: what decides a usable shot

AI Chronicle

Generation is short and cheap: Gen-4.5 makes 2 to 10 seconds at 720p, 24 or 25 frames per second, in six aspect ratios from 21:9 to 9:16.[3] Editing is long and expensive: Aleph 2.0 accepts 2 to 30 seconds of footage between 480p and 1080p, at 24 to 30 frames per second, with no more than 10 cuts. Clips over 30 seconds are trimmed, and you pay only for the range you edit.[6] Act-Two runs up to 30 seconds.[7] The Upscale app takes videos up to 40 seconds to 4K.[14]

Longer sequences are therefore assembled from several generations; Runway's image-to-video guide suggests using a clip's last frame as the next clip's input.[15] Each new generation is a chance for a face to drift, which our guide to keeping characters consistent in AI video addresses. Native output is 720p, and an upscaler adds pixels, not detail the generation never had.

Gen-4.5's help page lists no audio output.[3] Runway sells music, voice and sound effects as separate tools.[1] Plan dialogue as its own track; our AI voiceover workflow covers that layer.

The editing tools

Edit Studio (Aleph 2.0). You pick the frame that best shows what should change, describe the change and preview it as a still (5 to 20 credits, depending on the image model) before spending video credits. An optional "Extra motion" prompt adds movement the source lacks. Runway lists product swaps, relighting, background and wardrobe changes, and object removal as uses.[6]

Expand. Edit Studio can change aspect ratio by generating new picture at the edges. Runway warns: "Multi-shot videos may not work with this specific feature."[16] The main Edit Studio page still calls Expand "coming soon", though its own guide describes it as available.[6][16]

Keyframes. First-and-last-frame control now lives in the Animate with Keyframes app.[8][14] We found no spec page giving its duration, model or cost.

References. The References app takes up to three images plus a prompt to hold a character, product or style.[14] Our character consistency guide covers how to prepare them.

Act-Two. A driving performance (one person, face visible, waist-up at most, no cuts) animates a character image or video. Lower facial expressiveness "may improve character consistency in certain cases."[7]

Finishing tools include 4K upscaling, Ruby for SDR-to-HDR at 20 credits a second, and ProRes or PNG-sequence output on Max and Enterprise.[1][2][3] The Team page lists plugins for Premiere Pro, After Effects and Resolve.[12]

Ownership, commercial use and training

Runway's terms of use, updated 15 September 2026, answer three separate questions.[17]

Ownership. The terms say: "The Company does not claim ownership of any of your Inputs or Outputs."[17] The help centre adds that you keep ownership "as between you and Runway", and that commercial uses such as monetised YouTube, ads and festival entries are allowed without crediting Runway.[18] Free-plan videos carry a Runway watermark; paid plans remove it.[11]

Training. The same clause continues:

“You acknowledge that Inputs and Outputs may be used by the Company to train and improve its AI models”

— Runway, Terms of Use, section 4.4 (last updated 15 September 2026).[17]

The licence granted for that is perpetual, irrevocable and sublicensable, and we found no individual-plan opt-out in the terms or privacy policy. The no-training commitment is an Enterprise term; Runway Dev says it covers "your inputs and outputs on every model".[12][19] For unreleased products or client footage, that decides the plan.

Copyright. Runway not claiming your output is not the same as you owning a copyright in it. The US Copyright Office's position is that AI-assisted work can be protected where a human determines the expressive elements, but:

“Prompts alone, however, at this stage are unlikely to satisfy those requirements.”

— US Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (January 2025).[20]

Our explainer on AI training and copyright sets out where that line sits. The terms also warn that similar outputs may be generated for other users.[17]

Likeness. Runway's usage policy prohibits using "an image, video, or audio of another person without their permission."[21] An Act-Two performer or a recognisable face needs a release.

Provenance: C2PA

Runway's safety page states: "We implement C2PA so that AI-generated content can be traced back to its source." It also runs a public C2PA verifier.[22][23] The documentation does not say which export formats carry a manifest, so check your own files, and remember that re-encoding can strip the metadata. Our guide to content credentials explains what a manifest can and cannot prove.

Enterprise and the API

Enterprise adds SSO, roles, sharing policies, control over third-party models, legal indemnification and the no-training term.[12] The data-security page lists SOC 2 Type II and "technical separation between customer content storage and model training systems".[24] Runway Dev's Enterprise page offers "uncapped IP indemnification".[19] Read the carve-outs before relying on it. The API exposes Runway's four models and dozens of third-party ones at $0.01 a credit, but not the Apps.[2][13][14]

The lawsuit: Andersen v. Stability AI

Runway is a defendant in Andersen v. Stability AI, a proposed class action by visual artists in the Northern District of California (3:23-cv-00201). The claims concern Stable Diffusion, an image model, not Runway's video models. The artists allege that Runway "worked with, helped train, and then distributed Stable Diffusion with Stability AI."[25] These are allegations, not findings.

On 12 August 2024, Judge William Orrick denied Runway's motion to dismiss the copyright infringement claims, dismissed the DMCA claims with prejudice and the unjust-enrichment claim with leave to amend.[25] Runway had not moved to dismiss the training-based direct-infringement claim. Letting claims proceed is not a finding that anyone infringed or that fair use fails.

A February 2026 order set trial for 1 June 2027.[26] On 28 September 2026 the court granted the artists more time: class certification and summary judgment will be heard by 29 September 2027, and trial is "To be set by Court."[27] The order adds:

“I do not expect to grant any further continuances.”

— Judge William H. Orrick, order modifying the case schedule, ECF No. 764 (28 September 2026).[27]

Some websites reported a trial starting in September 2026; the docket shows none has. For customers, the case changes no current terms, but it is a reason to know which plan carries indemnification.

Documented limits and risks

• Physics and causality. Runway's Gen-4.5 announcement lists limitations "common to AI video generation models": effects that "sometimes precede causes", objects that "disappear or appear unexpectedly", and "success bias", where actions succeed too often.[9]
• Short, 720p generations: 10 seconds maximum for Gen-4.5.[3]
• Edit constraints: Aleph 2.0 rejects clips with 11 or more cuts; Expand prefers single shots.[6][16]
• Credit burn: Standard and Pro credits expire monthly; failed takes cost the same.[10]
• Training by default outside Enterprise.[17][12]
• A moving product: the free plan's models "can change", and the documentation is not always in step, as the Turbo and Expand conflicts show.[11]

Who it suits, and who should skip it

It suits editors and small studios who want to change footage they already have: product swaps, seasonal versions, relighting, reframing for vertical. Aleph 2.0's 30-second window and preview step are built for that. It also suits teams wanting one credit pool across several vendors' models, and creators who need performance capture without a mocap rig.

Consider Enterprise, or skip, if you handle unreleased products or client footage you cannot let train a model. Skip too if you need long single takes with dialogue from one generation: Runway's own model makes up to 10 silent seconds at 720p.

Free is for looking, not publishing. The 125 one-time credits work only with a limited set of models: Gen-4.5 and Act-Two require Standard or higher, and every free video carries a watermark.[3][7][11]

How to test it yourself

1. Start on Standard for one month and note your starting credit balance.
2. Write a five-shot list from a real project: two generated shots, two edits of your own footage, one Expand to 9:16.
3. Draft in Gen-4 Turbo, finish in Gen-4.5, as Runway suggests. Log credits spent per kept second for each.
4. Edit one of your own clips in Edit Studio. Preview the keyframe, apply one change, and check frame by frame that nothing else moved.
5. Run Expand on a single-shot clip and a clip with a cut. Compare the edges and note any artefacts.
6. Check provenance. Download the final file, then run it and a re-exported copy through Runway's C2PA verifier.
7. Read the clauses that matter to you: section 4.4 of the terms on training, and your plan's watermark and rollover rules.

A worked example (hypothetical): budgeting a seasonal ad refresh

This is an illustrative scenario. The business, footage and attempt counts are invented to show how Runway's published rates add up. We did not run these edits, and nothing here reports a result.

A small skincare brand has a 25-second hero ad shot on camera. It wants three seasonal versions (winter, spring, holiday), each with a vertical 9:16 cut, plus two new 5-second product close-ups.

That is about three Pro allowances, most of a Max month (9,500) or one Team seat (6,900). Editing dominates: 25 seconds of Aleph 2.0 costs 700 credits per attempt, more than Standard's whole month. Two documented details help. Edits are charged only for the range you select.[6] And because Expand prefers single shots, the ad should be expanded shot by shot before the cuts are assembled.[16] With unreleased product footage, the brand must also decide whether training on its inputs is acceptable.

Questions readers commonly ask

Can I use Runway videos commercially?

Yes on Runway's side: it claims no ownership and does not restrict commercial use of outputs, subject to its terms.[17][18] You still need rights to your inputs and releases for any real person, and in the US a purely prompted output may not be protected by copyright.[20][21]

Does Runway train on my uploads?

Under the standard terms, inputs and outputs "may be used" to train its models. Enterprise agreements commit not to train on your content.[17][12][19]

Does the lawsuit affect my use of Runway?

Not directly. Andersen v. Stability AI concerns Stable Diffusion's training and distribution, has no merits ruling, and has no trial date.[25][27]

The next step

Runway's tools reward planning: short shots, fixed references, edits applied to footage you control. The hardest of those to get right is a character who looks the same from shot to shot. Continue with how to keep characters consistent in AI video.

Sources and evidence note

This review rests on Runway's pricing page, help centre, Academy, developer documentation, terms, policies and research announcements, a US Copyright Office report, and the federal court docket, all accessed on 16 October 2026. Vendor documents establish what Runway says about its own products, not how well they perform. Prices and model menus change often; every figure is dated. Per-second dollar costs and credit budgets are AI Chronicle's arithmetic from published rates. The worked example is hypothetical. No hands-on test is claimed.

Related reading

• ElevenLabs review: voice production and editing
• Best AI image generators for creator workflows