Runway is an AI research company whose products generate and edit video and images through models such as Gen-4.5, Aleph and Gen-4 Image, billed by a shared credit pool. It serves filmmakers, marketing teams and developers, and has expanded beyond media into general world models for gaming and robotics.
Used this tool? Rate it
Used this tool? Rate it
Runway is an AI research company that builds generative models for video and images, and increasingly for simulating the physical world. Its current positioning is stated on the homepage: "Runway is building foundational Real-World Intelligence that can understand, simulate and act in the world."
That sentence signals something important for anyone who last looked at this product a year or two ago. Runway was widely known as an AI video generator, and it still is one — but the company now frames itself primarily as a world-model research lab whose creative tools are one application among several. The About page puts the thesis plainly: "We believe that world models are at the frontier of progress in artificial intelligence." Alongside its video products, the homepage now features GWM-1, described as a general world model built to interact with the real world, and the product line is split into three parallel platforms: Runway Creative, Runway Dev and Runway Robotics.
One practical note before anything else. This entry's registered URL is runwayml.com, which now issues a permanent redirect to runway.com. The older domain still works as an entry point, but the canonical brand domain has changed, and the legal pages live at /terms-of-use and /privacy-policy rather than the more commonly assumed /terms and /privacy.
The operating entity is disclosed clearly: the operating entity is Runway AI, Inc., with a principal address in Dover, Delaware, and the terms are governed by New York law. The company reports offices in New York, San Francisco, Seattle, London, Tel Aviv and Tokyo, plus remote staff.
On scale, this is not a small startup and the figures come from independent reporting rather than vendor claims. TechCrunch reported in February 2026 a $315 million round led by General Atlantic at a $5.3 billion valuation, with participation from Nvidia, Adobe Ventures and AMD Ventures among others. The company was founded in 2018 in New York by Cristóbal Valenzuela, Anastasis Germanidis and Alejandro Matamala-Ortiz, and earlier raised a $308 million round in April 2025.
Gen-4.5 is the flagship. Runway describes it as "The world's best video model, featuring state-of-the-art motion quality, prompt adherence and visual fidelity." That is a vendor claim and should be read as such — no independent benchmark is cited — but the three axes named are the right ones to evaluate: whether motion looks physically plausible, whether the output matches the prompt, and whether individual frames hold up. The platform also exposes third-party models, with Seedance 2.0 Pro and Seedance 2.0 Fast appearing as separately priced options.
Aleph 2.0 addresses a different problem from generation: modifying video that already exists. This matters because generating an entirely new clip is often the wrong tool when you have usable footage that needs a change. Editing existing material is also where prompt-driven tools tend to be most useful in real production, since the composition is already fixed.
Images are a first-class capability, not an afterthought, with their own pricing. Gen-4 Image and its Turbo variant are separately metered, and image generation is dramatically cheaper than video — relevant both for producing stills and for creating reference frames to drive video generation.
Higher tiers add custom voice creation for lip sync and text to speech, and 4K upscaling is available from the Standard tier upward. Act-Two, listed under research, addresses performance capture — driving a generated character from recorded human performance.
GWM-1 represents the strategic direction. Rather than producing footage for viewing, a world model aims to simulate an environment that an agent can act within, which is why robotics and gaming are the named target industries. TechCrunch's reporting notes Runway is "increasingly seeing adoption in gaming and robotics" beyond its traditional media customers. For most creative users this has no immediate effect, but it explains where the company's investment is going.
Pre-visualisation and concept work in film and advertising is the most established use. Generating a rough version of a shot before committing budget is cheaper than shooting it, and quality that would be marginal for final delivery is entirely sufficient for demonstrating intent.
Marketing and social video production is the highest-volume use. Teams needing many short pieces benefit from generating and iterating quickly, especially when combining generated footage with edits of existing material.
Editing and salvage work is often underestimated. Using Aleph to alter footage you already have solves a common production problem — needing a variant of a shot without a reshoot.
Independent filmmaking and music video work draws users who need visuals that would be impossible on a small budget, where a distinctly synthetic aesthetic is acceptable or even desirable.
Developer and API integration is a distinct audience served by Runway Dev, embedding generation into other products rather than using the web interface.
Robotics and simulation is emerging rather than established, and prospective users there should treat it as a research direction rather than a mature product.
Budget in credits, not dollars. The subscription price tells you little; what matters is how many seconds of the specific model you need. Do that arithmetic against your actual workload before choosing a tier.
Prototype in images. Because image generation costs a small fraction of video, resolving composition and style as stills first is the most effective way to reduce wasted spend.
Match the model to the task. Attempting to regenerate a whole scene when you only need to change one element is both more expensive and less reliable than editing the footage you have.
Read the training clause before uploading client work. Runway's terms permit using inputs and outputs to train its models. For confidential or contractual material this is a genuine consideration and may require client consent.
Do not assume commercial safety equals legal safety. The terms permit commercial use of outputs, but that does not resolve third-party rights in likenesses, trademarks or recognisable people appearing in your generations.
Expect the product to change. This company ships frequently — four significant releases between April 2025 and May 2026 alone — so verify pricing, credit rates and model availability at the source rather than trusting any secondary description, including this one.
Get consent before using someone's likeness or voice with lip sync and custom voice features, which is an obligation no software setting resolves for you.
Film and video professionals doing pre-visualisation, concept work or effects experimentation are the core audience, and the tolerance for imperfect output is highest here.
Marketing and content teams producing high volumes of short-form video benefit from speed and iteration more than from frame-perfect fidelity.
Independent creators and small studios gain access to visual effects that would otherwise be out of reach financially.
Developers integrating generation into their own products are served by the API platform.
It is a poor fit for anyone needing guaranteed frame-accurate control, for productions where a synthetic look is unacceptable, for teams with strict confidentiality requirements that conflict with the training clause, and for casual users who only occasionally need a single clip — for whom the credit-based subscription is poor value.
Runway is delivered primarily as a browser-based application, with generation running on Runway's infrastructure rather than local hardware, so no powerful GPU is required.
The product surface is organised into three platforms: Runway Creative, Runway Dev and Runway Robotics — the creative tools, the developer API, and the robotics and world-model work respectively. The developer platform is the route for embedding generation into other software.
Storage is allocated by plan, from 5GB on the free tier up to 500GB on Pro, with more on higher tiers. Because processing is server-side, a reliable connection matters more than local machine specifications, and large uploads and downloads are a practical consideration for high-resolution work.
Runway uses credit-based pricing across five tiers, and understanding the credits matters far more than comparing the headline subscription prices.
The Free plan gives 125 one-time credits to explore Runway's AI tools, plus 5GB of storage, with watermarked output.
The paid tiers, at annual billing rates, are Standard at $12/month with 625 credits/mo, Pro at $28/month with 2250 credits/mo, and Max at $76/month with 9500 credits/mo. Monthly billing raises these to $15, $35 and $95 respectively — a 20 percent difference. All paid tiers remove watermarks; Standard adds 4K upscaling, Pro adds custom voices for lip sync and text to speech plus 500GB of storage, and Max allows unused credits to roll over for one month. Enterprise is quoted individually.
The consumption rates are published and are the figures that actually determine cost. Gen-4.5 consumes 60 credits per 5 seconds, Aleph 2.0 costs 140 credits per 5 seconds, Seedance 2.0 Pro runs 160 credits per 4 seconds at 1080p, and Seedance 2.0 Fast 116 credits per 4 seconds. Images are far cheaper: Gen-4 Image costs 8 credits per image at 1080p and just 2 credits on the Turbo variant.
These rates are worth checking arithmetically, and they hold up. Runway also publishes annual generation counts per tier, and they reconcile exactly: Standard's 7,500 annual credits yield 125 Gen-4.5 videos, Pro's 27,000 yield 450, and Max's 114,000 yield 1,900. The same consistency holds for Aleph and both Seedance variants.
The practical implication is stark. On the Standard plan, a full month's credits produce roughly 52 seconds of Gen-4.5 video. Anyone planning sustained video work should calculate required seconds per month against these rates before selecting a tier — and should re-check the numbers on the pricing page, since models and rates change frequently.
OpenAI's Sora and Google's Veo are the closest frontier-model comparisons, differing mainly in access model and ecosystem rather than in fundamental capability.
Kling and Hailuo compete directly on video generation, often at lower cost, and are worth benchmarking on your own prompts since relative quality shifts with each release.
Pika and Luma target similar creative users with different interface philosophies and pricing.
Adobe Firefly appeals to teams already in the Adobe ecosystem and prioritising commercially cleared training data.
For editing rather than generating, conventional tools such as Premiere Pro or DaVinci Resolve remain far better for anything requiring precise control, and many practitioners combine them with Runway rather than choosing between them.
The training clause deserves the most attention. Runway's terms state that "Inputs and Outputs may be used by the Company to train and improve its AI models", with users granting a perpetual, irrevocable, sublicensable licence for that purpose. This coexists with a favourable ownership position, and both halves must be read together: you own your outputs, and the company may still learn from them. For agencies handling confidential client material, this needs checking against client agreements.
Credit economics constrain real use more than any feature limitation. At 60 credits per 5 seconds for Gen-4.5, the Standard tier's monthly allowance covers well under a minute of flagship-model video. Costs rise sharply with the more expensive models.
The vendor's quality claim is unverified. Describing Gen-4.5 as "the world's best video model" is marketing language with no independent benchmark cited, and comparative quality in this field changes with nearly every release cycle.
Independent review coverage is thin relative to the company's prominence. On Product Hunt the product holds a 4.4 out of 5 rating from just 18 reviews — a small sample that should be reported with its size rather than treated as authoritative. G2 could not be verified: both direct fetch and a fully rendered browser session returned HTTP 403, so no rating from that source is asserted here.
Name collision is a real hazard in research. A separate Product Hunt entry for a product called Runway describes a startup cash-planning tool with no relation to this company; anyone gathering third-party data should confirm which Runway a listing refers to.
The minimum age is 13, with parental consent required under 18 — permissive relative to the content generation capability, and worth noting for institutional deployments.
Finally, generative video retains well-known technical limits: physical implausibility in complex motion, difficulty with consistent text and hands, and identity drift across shots. Precise creative control remains weaker than in conventional tools, which is why generation and traditional editing are usually complements rather than substitutes.
Data collection is broad, as expected for a generative platform. The privacy policy covers identifiers, account data and user content including "prompts, photos, images, music, videos, audio, screen sharing, comments, questions, messages, works of authorship". Most significantly, it also covers biometric data: "voice data and scans of faces that you submit when you use certain product features", which applies to lip sync, custom voices and performance capture.
Biometric retention is bounded with an explicit ceiling. The policy commits to retaining face scans and voiceprints for "no longer than the earlier of the date the initial purpose for collecting the data is satisfied or within three years of your last interaction with us." A stated numeric limit on biometric retention is better disclosure than many competitors provide.
Regulatory rights are comprehensive, covering GDPR entitlements — access, deletion, correction, restriction, objection, portability and complaints to a supervisory authority — alongside CCPA rights including the right to know, delete, correct and opt out of sale or sharing. Named third parties include Amazon Web Services, Stripe and Google Analytics.
The genuinely important interaction is between the privacy policy and the terms of use. The privacy policy commits that biometric data is used only to provide the requested service, while the terms permit inputs and outputs generally to train Runway's models. Those are consistent but distinct commitments, and users concerned about content confidentiality should read the terms rather than the privacy policy for the answer that affects them.
On ownership, the position is clear and favourable: "Company does not claim ownership of any of your Inputs or Outputs." Commercial use of outputs is not restricted, subject to compliance with the agreement — but as noted above, that resolves the platform's claims, not third-party rights.
There is a free tier with 125 one-time credits and 5GB of storage, but outputs carry a watermark and the credits are a single allocation rather than a monthly refresh. It is an evaluation tier: enough to judge output quality on your own material, not enough for ongoing work. Removing watermarks requires a paid plan starting at $12 per month billed annually.
Billed annually, Standard is $12 per month with 625 credits monthly, Pro is $28 with 2,250 credits, and Max is $76 with 9,500 credits. Monthly billing costs $15, $35 and $95 respectively. Enterprise is priced individually. Compare tiers by credits needed, not by subscription price.
Rates are published per model. Gen-4.5 costs 60 credits per 5 seconds, Aleph 2.0 costs 140 credits per 5 seconds, Seedance 2.0 Pro costs 160 credits per 4 seconds at 1080p, and Seedance 2.0 Fast costs 116 per 4 seconds. Images are far cheaper at 8 credits, or 2 on Turbo. In practice the Standard plan's monthly credits produce roughly 52 seconds of Gen-4.5 video. Verify current rates on the pricing page, as they change with model releases.
Yes, subject to compliance with the agreement. Runway does not claim ownership of your inputs or outputs and does not restrict commercial use of outputs. However, this does not clear third-party rights: recognisable people, trademarks, licensed music or copyrighted source material still require their own permissions.
The terms state that inputs and outputs may be used to train and improve Runway's AI models, and using the platform grants a perpetual, irrevocable, sublicensable licence for that purpose. If you handle confidential or client-owned material, review this clause against your obligations before uploading.
The privacy policy treats face scans and voice data as biometric information collected when you use features such as lip sync or custom voices, and commits to using it only to provide the requested service. Retention is capped at the earlier of the purpose being fulfilled or three years after your last interaction.
Gen-4.5 generates new video from prompts or images and is the flagship model for creating footage that does not exist. Aleph 2.0 edits and transforms video you already have. Aleph costs substantially more per second, so choose based on whether you need creation or modification.
The company has moved to runway.com as its primary domain, and runwayml.com now issues a permanent redirect there. Both reach the same product. Note that the legal pages are at /terms-of-use and /privacy-policy rather than the shorter paths people often assume.
No longer, and this is the most significant recent change. The company positions itself as a world-model research lab, with GWM-1 presented alongside its video models and the product line split into Runway Creative, Runway Dev and Runway Robotics. Independent reporting notes growing adoption in gaming and robotics beyond media customers. The creative video tools remain fully available and are still what most users come for.
Treat them cautiously. Runway's description of Gen-4.5 as the world's best video model is a vendor claim with no independent benchmark cited, and leadership in this field changes frequently. Third-party validation is limited: 4.4 from 18 Product Hunt reviews is a small sample, and G2 could not be accessed for verification. Test on your own material before committing to a plan.
