Runway Review 2026: Pricing, Unlimited Mode & Agent

Is Runway worth it in 2026?

Runway is worth it when the hard part is generating or transforming shots, building campaign variations, directing several media models from one Agent conversation, or assembling a short sequence around those assets. Its economics changed materially in August: a new Unlimited Mode is available on Pro and Max for supported models without spending credits, at the cost of slower starts and lower concurrency. That feature is different from the legacy Unlimited subscription and Explore Mode, which Runway says will retire on September 1, 2026. Credits still matter for unsupported models, faster Credits Mode, and the rest of the catalog. Runway also has a real multi-track Studio timeline, but it does not accept images on that timeline and remains shallower than a mature long-form NLE. Choose it for generative production and shot-level transformation; benchmark it before making it the only system for long, faithful source-footage editing.

How much does Runway cost in 2026?

Runway charges per paid editor while each workspace shares one monthly plan-credit pool. Annual plans are prepaid for the year, but credits still arrive and reset monthly. Pro and Max can also expose a separate Unlimited Mode for supported models; it uses no credits, starts more slowly, and allows fewer concurrent jobs. Prices and visible allowances below were checked on the live US pricing page August 26, 2026; tax, currency, promotions, model eligibility, and Enterprise quotes can differ.

Included-credit economics remain $0.024 per Standard credit monthly or $0.0192 annually, about $0.0156/$0.0124 for Pro, and $0.01/$0.008 for Max. Purchased web credits start at 1,000 for $10, do not expire, and are separate from developer API credits. Unlimited Mode changes marginal cost only for supported operations: it does not spend credits, quality is the same as Credits Mode, but starts are slower and concurrency is lower. The live matrix is model-specific. Examples checked August 26 include Gen-4.5 at 60 credits/5s, Aleph 2.0 at 140/5s, Gen-4 Turbo at 25/5s, Kling 3.0 Standard with audio at 65/5s, Kling 3.0 Pro with audio at 85/5s, Veo 3.1 with audio at 320/8s, Nano Banana Pro 2K at 20/image, Seed Audio at 15/minute, and text-to-speech at 1/50 characters. Confirm the exact model, mode, and cost immediately before generating.

Runway at a Glance

What Runway Actually Is

The simplest accurate description of Runway is a creative operating system for generated media. It brings first-party and selected third-party image, video, and audio models into a shared asset library, then offers several ways to operate them: direct Tool Mode, task-specific Apps, a conversational Agent, reusable node-based Workflows, an MCP connection for external assistants, and a separate developer API. A buyer is not purchasing one deterministic video editor. They are purchasing access to a changing catalog of probabilistic creative operations, with credits as the common meter inside the web app.

Runway Agent is the broadest interface. A user can supply a product page, image, existing ad, document, campaign result, or plain-language idea and ask Agent to build marketing assets, product videos, social variations, image ads, voiceover, music, or a campaign. It analyzes the request, proposes a plan, chooses capable models, generates media, and accepts revisions in the same conversation. Runway says the plan describes intent rather than a guaranteed result. That qualification matters: Agent can make mistakes, model outputs can drift, and a user should expect evaluation and iteration rather than a one-command final deliverable.

The historical distinction is still important. Runway retired Video Editor Projects, Inpainting, and the old Remove Background workflow on July 30, 2026. Its current help points former Video Editor Projects users to Timeline Studio and routes Inpainting or Remove Background work toward Edit Studio. Older reviews that judge the current product from the retired editor are stale; reviews that claim Runway has no timeline are also stale. The fair conclusion is narrower: Runway restored useful assembly inside its generative system, but its strategic center is generative creation and transformation rather than deep long-form post-production.

Edit Studio is the current existing-footage transformation workflow, powered by Aleph 2.0. It can swap a product, character, background, or object; remove an element; add visual effects; alter lighting, weather, time of day, wardrobe, or style; and propagate a changed reference frame across a short clip. The user uploads video, optionally provides a reference image, chooses or generates a keyframe, describes the intended change, previews that frame, and generates the selected video range. This is semantic pixel editing: it changes what the shot appears to contain, not merely where a clip sits on a timeline.

Edit Studio has unusually concrete boundaries. A source must be longer than two seconds and shorter than 30 seconds; footage over 30 seconds is automatically trimmed. Inputs must be 480p through 1080p, use 24–30 FPS, have a conventional aspect ratio, and contain no more than ten cuts. Higher frame rates are downsampled. The current interface exposes Single edit, while Multi-edit and Expand are labeled as coming soon. By default the operation covers the full clip, but range handles can limit both the changed region and the credit charge. Those specifications make it excellent for individual shots and unsuitable for an uninterrupted hour-long edit.

Output choice also depends on surface and plan. MP4 is the default. Max, legacy Unlimited, and Enterprise users can select ProRes MOV or a PNG image sequence at generation time in Tool Mode or Edit Studio for an additional five credits per second. ProRes supports profiles from 422 Proxy through 4444 XQ and includes PCM audio; PNG sequences arrive as ZIP archives. Agent and Workflows outputs remain MP4, and an existing MP4 cannot simply be converted into a premium sequence without regenerating. Range edits are MP4 even when the broader plan supports studio formats.

Gen-4.5 itself is a short-shot generator, not a long-form renderer. The official guide lists text-to-video and image-to-video, 2–10-second duration, 720p output, 24 or 25 FPS, and several landscape, portrait, square, and cinematic aspect ratios. Standard or higher is required. Max, legacy Unlimited, and Enterprise can add ProRes or PNG sequence output for five extra credits per second. Runway's own guidance emphasizes complex sequenced prompts, camera choreography, scene composition, event timing, and atmospheric change. That is a compelling filmmaking primitive; it is not a promise that a complete scene will emerge correctly on the first attempt.

Apps package common operations into narrower interfaces. The current catalog includes Edit Studio, Performance Capture with Act-Two, Backdrop, Character Swap, Color Grade, Lighting, Motion Sketch, Multi-Shot, Product Shot Video Builder, Remove, Stitch Video, Stylize Video, Time of Day, Upscale, Weather, ad and product-image tools, and audio utilities. Stitch Video can combine clips up to a total of 60 minutes, while the Upscale app accepts video up to 40 seconds for a 4K output. Apps reduce prompt and node setup, but each App has its own accepted inputs and credit cost.

Act-Two illustrates the difference between an editing feature and a conventional edit. It transfers motion, speech, facial expression, and—when an image is used—gestures from a driving performance into a character. It supports up to 30 seconds at five credits per second. A character video preserves its camera and environmental motion while driving facial performance; a character image can support broader gesture control and automatically added environmental motion. The creative result can be powerful, but source choice, face visibility, lighting, framing, motion, and consent remain the operator's responsibility.

Workflows are Runway's reusable automation layer. A node canvas can chain input media, prompts, language models, image models, video models, audio operations, and utilities. Users can test individual nodes or run the graph, start from templates, reuse a process, and share it. Agent can now propose, create, discover, edit, and run Workflows from chat. Building the graph itself does not consume credits; running credit-bearing nodes does. With Ask before generating enabled, Agent previews the proposed models, prompts, and estimated run cost before executing a multi-node workflow.

Runway MCP serves a different automation audience. It connects Runway to MCP-capable products such as Claude, ChatGPT, Cursor, Replit, and other clients at Runway's remote MCP endpoint. Authentication uses a Runway sign-in rather than manually copied API keys. An external assistant can request images, video, product-URL marketing content, or dialogue ads, and the resulting media appears in the Runway library. MCP uses the same web-app credit balance. It is convenient for individual assistant workflows, but it is not the same product or balance as the separate production developer API.

Team economics require a second calculation. Standard supports up to five paid editors; Pro and Max support up to ten. Each Admin or Editor is charged the full seat price, yet a workspace receives only one shared plan-credit allowance. Two Standard editors therefore cost $30 monthly or $288 annually but still share 625 credits per month—not 1,250. Viewers are free. Added paid seats are prorated. Removing a member does not create a cash refund; remaining seat value is handled as invoice credit and an unused seat may be reassigned. Enterprise adds broader role, analytics, SSO, and organization controls.

Billing is conventional SaaS but worth reading before annual purchase. Plans auto-renew. Upgrades take effect immediately with proration, while downgrades generally create invoice credit. Annual-to-monthly changes can leave an account credit unless support approves a different resolution. Canceling means downgrading to Free; paid access remains until the period ends. Assets and projects remain unless the account is deleted, although a Free account above its allowed project count may be unable to create more. A cancellation must be completed through the final Stripe confirmation step, and each workspace can carry its own subscription.

Runway's refund language is discretionary rather than a broad guarantee. The help center says a plan payment may qualify within 30 days when the plan is unused or close to unused, unused purchased credits may qualify within 30 days, and a mistaken plan or cadence may qualify for the difference. An approved plan refund ends paid access immediately and removes unused plan credits. A generation that completes but looks wrong, fails to preserve a likeness, or disappoints the user is not automatically refunded; only a generation error should restore credits automatically.

Content rights and data use must be separated. The self-serve Terms say Runway does not claim ownership of inputs or outputs and does not restrict compliant commercial use of outputs. They also say outputs may not be unique, accuracy or reliability is not guaranteed, and the user is responsible for having all necessary rights. Most importantly, the same Terms permit Runway to use inputs and outputs to train and improve AI models, algorithms, products, and services under a broad perpetual license. The Enterprise page takes a materially different position: it advertises that Runway does not train on Enterprise data and that customers own their outputs. A sensitive buyer should obtain the applicable promise in the signed contract rather than infer Enterprise protection from a self-serve plan.

Privacy and security are also different questions. Current asset-sharing help says assets are private by default and become available only when intentionally shared. An anyone-with-link asset can be viewed and downloaded, and a generated asset's shared page can expose prompts, input details, seed values, and other generation metadata. Runway states SOC 2 Type II, TLS 1.2 or newer in transit, AES-256 at rest, role-based production access, backups, incident handling, and a deletion framework. Its Privacy Policy uses purpose-based general retention rather than one simple deletion deadline, allows processing in the United States and other countries, and sets a special biometric retention ceiling for face scans and voiceprints.

The purchasing decision follows the unit of work. Runway is strongest when the unit is a shot, asset family, product campaign, visual effect, character performance, or reusable generative pipeline. It is weaker when the unit is a faithful hour-long recording that needs editorial judgment across a transcript, exact multicamera decisions, detailed sound mixing, and a controlled final timeline. The current Agent timeline narrows that gap for basic assembly, but it does not erase it. Test the exact source material, the exact model, and the complete iteration budget before standardizing a production workflow.

Studio is the current name to remember for timeline assembly. A user can begin manually or ask Agent to construct a sequence, drag video into the timeline, upload reference media in chat, add tracks and files, split or trim clips to roughly one tenth of a second, move clips between tracks, multi-select them, delete them, detach audio to its own track, and run Detect Shots to create cuts at scene boundaries. The final sequence can then be downloaded. This is enough for a short social assembly or generated campaign rough cut; it is more than a gallery of disconnected generations.

Studio also publishes limits that define what it is not. Images are not currently supported on the timeline. Runway advises against very large or long files. Its guide documents clip arrangement, shot detection, splitting, trimming, audio separation, tracks, and download—not the mature multicamera switching, transcript editing, caption delivery, audio buses, keyframe automation, color management, plugin ecosystem, relinking, interchange, and long-program conform expected from a full NLE. Calling it a real timeline is accurate; calling it a complete post-production replacement is not.

Unlimited Mode is the largest economic change in this review. On eligible Pro and Max workflows, a mode switch can run supported models without subtracting credits. Runway says output quality is identical to Credits Mode. The trade is scheduling: an Unlimited job may wait longer to start and fewer jobs can run at once. If the concurrency allowance is full, Generate can remain disabled until an active job completes. A buyer should therefore measure time to first frame, total queue time, and approved output per hour—not just credit consumption.

The word unlimited needs four qualifications. It applies only to supported models and tools; availability can vary by plan or promotion; the toggle is hidden when the selected operation is unsupported; and the platform still enforces concurrency. Credits Mode remains available for faster starts and a broader supported catalog. Unlimited Mode lowers marginal generation cost for patient iteration, but it is not an SLA, an unlimited parallel queue, or a promise that every newly added model will be included.

The new Unlimited Mode is distinct from the legacy Unlimited subscription. The old plan included 2,250 monthly credits plus Explore Mode. Runway launched Max with 9,500 credits, one-month rollover, first access, and premium production features, and says existing monthly Unlimited plans will become Max at the same $95 price on September 1, 2026. Eligible annual users in August were offered either a refund of remaining months or a migration bonus. New buyers should evaluate the current Pro/Max Unlimited Mode documentation, not assume legacy Explore entitlements survive unchanged.

Current model pricing spans more than one order of magnitude. Five seconds cost 25 credits on Gen-4 Turbo, 60 on Gen-4.5, 65 on Kling 3.0 Standard with audio, 85 on Kling 3.0 Pro with audio, and 140 on Aleph 2.0. Eight seconds of Veo 3 or Veo 3.1 with audio costs 320 credits. Other visible rows include Kling 3.0 Turbo and Motion Control, Seedance 2.0 variants, Kling O3 and 2.x models, WAN 2.6, HappyHorse, Grok Imagine, Act-Two, and first- and third-party image or audio tools. A plan allowance is therefore meaningless without a model-and-duration workload.

The live matrix also exposes why old examples can mislead. The Standard plan guide includes an example described as four Video-to-Video generations with Aleph for 60 credits, while the current price page lists Aleph 2.0 at 140 credits for five seconds. Those statements need not describe the same model generation: Gen-4 Aleph and Aleph 2.0 appear as separate priced rows. Never convert a generic ‘Aleph’ example into Aleph 2.0 capacity. Record the full model label, version, duration, resolution, audio setting, and mode shown at confirmation.

Seedance 2.5 is a useful stress test for credit math because it accepts unusually rich references. Runway documents text, image, video, and audio inputs; as many as 50 total references composed of up to 30 images, 10 videos of no more than 30 seconds each, and 10 audio files; and Reference, Keyframe, Edit, and Extend workflows. Outputs can run 4–30 seconds or Auto across common landscape, portrait, square, and 21:9 ratios at 480p, 720p, or 1080p in MP4 or MOV.

Seedance 2.5 charges both for output and, optionally, video references. Base rates are 68 credits per second at 1080p, 30 at 720p, and 20 at 480p. An input video can add 34, 15, or 10 credits per referenced second at those resolutions. A 30-second 1080p output is therefore 2,040 credits before video references; one 30-second video reference can add 1,020, raising the total to 3,060. The same 30-second output is 900 credits at 720p or 600 at 480p before reference charges.

Those numbers make plan fit concrete. Standard's 625 credits cannot cover even a ten-second 1080p Seedance 2.5 output at the documented 680-credit base rate, although it can cover 30 seconds at 480p with 25 credits left. Pro's 2,250 can cover one 30-second 1080p output without a video-reference surcharge and retain 210 credits. Max can cover four such base outputs for 8,160 credits and retain 1,340. Unlimited Mode can change that calculation only when Seedance 2.5 is actually supported and the user accepts its queue behavior; the confirmation surface is authoritative.

Seedance's four workflows have different continuity implications. Reference mode composes from supplied assets. Keyframe mode lets the creator define important frames. Edit mode matches the source video's aspect ratio and length. Extend also matches aspect ratio but returns only newly generated footage, so the operator must assemble it with the original clip. Sketch can anchor drawn guidance to a chosen video frame. These are powerful controls, yet 50 references do not guarantee identity, physics, typography, product fidelity, or continuity across a final sequence.

Agent preferences improve repeatability without becoming hard constraints. A user can state default model, resolution, aspect ratio, or task rules in natural language—for example, preferring Seedance 2.5 for longer videos—and save them as defaults. Runway explicitly treats those preferences as guidance. Agent can override them when the model cannot satisfy the task or the current plan lacks access. Production teams should capture the actually selected model and settings in output metadata instead of assuming a saved preference executed.

Sessions organize generations and conversations, but they are not the same as asset custody. Workspace members can collaborate and share session context. Deleting a session does not delete its generated assets, while a deleted session itself cannot be recovered. The asset library separately handles uploads, generated media, folders, downloads, and deletion. A durable workflow needs deliberate naming, external masters, rights records, and a retention policy across both surfaces.

Third-party models expand choice and the procurement surface. Enterprise administrators can enable or disable them. Runway says that when they are disabled, data is not sent to those providers, though features may become unavailable. For enabled Enterprise use, Runway says its terms and DPA cover providers, providers are contractually prohibited from training on customer content, Runway's indemnity extends to third-party models, critical vendors receive annual security review, and Seedance and Kling providers state that customer data is not processed in China. Those are useful first-party assurances that should still be reflected in the signed order, DPA, subprocessor review, and intended region.

Enterprise governance reaches beyond model toggles. Runway lists SSO, organization and workspace controls, analytics, spend notifications, the ability to disable external-link sharing, prompt metadata embedded in files, optional face-detection disabling for Act-Two images, Brand Kits, priority support, and negotiated legal terms. Some published features are marked for full-service arrangements rather than every Enterprise contract. Procurement should attach each required control to an entitlement, administrator test, evidence owner, and remedy rather than rely on the category name.

Operationally, the right benchmark has two lanes. In Credits Mode, measure credits per approved second, failed-generation restoration, top-up use, and monthly expiration. In Unlimited Mode, measure queue time, concurrency, supported-model coverage, completion rate, and operator waiting. Across both, measure prompt iterations, identity drift, temporal errors, typography errors, source-faithfulness, correction time, export compatibility, and approved deliverables per paid editor. A cheap generation that needs hours of rescue is not cheap production.

Runway's public status page is useful during an incident, but it is not a substitute for project-level observability. Record the session, model, mode, start time, completion time, credits shown before and after, error state, export ID, and support case. For client work, keep source media and approved masters outside the platform. That evidence distinguishes a product outage, a queue delay, a model limitation, a credit-accounting question, and an operator decision.

Where Runway Wins

Limits and Trade-offs

Who Should Choose Runway?

• Filmmakers, artists, and ad teams generating new shots or transforming individual source clips.
• Performance marketers producing coordinated image, video, voiceover, music, resize, and localization variants.
• Creative technologists turning successful prompt experiments into reusable node-based Workflows.
• Teams that want one workspace for first-party and selected third-party generative media models.
• VFX and previsualization work involving products, characters, backgrounds, lighting, weather, motion, or style.
• Creators who want an external MCP assistant to generate into the same Runway asset library.
• Organizations willing to negotiate Enterprise governance, SSO, analytics, custom credits, and no-training terms.
• Production stacks that already have a separate system for deep long-form editing and finishing.
• Pro or Max creators who can trade queue speed for no-credit iteration on currently supported Unlimited Mode models.
• Teams testing Seedance 2.5 with dense image, video, and audio references and explicit resolution-level budgets.
• Short-form producers who need shot detection, clip assembly, and audio separation around generated assets rather than a complete desktop NLE.
• Enterprise buyers who need administrator control over third-party models, external sharing, SSO, spend visibility, and contractual AI-data terms.

Runway vs Valmera

Runway Agent and Valmera both accept conversational direction, but the delegated job is different. Runway's center of gravity is generating media, semantically transforming short shots, coordinating campaign assets, and chaining generative models. Valmera's center is indexing real uploaded footage, changing a versioned edit decision list, rendering a preview, and revising the editorial result through conversation.

Runway is the stronger choice when the brief says invent a shot, replace a product, change the environment, animate a character, create multiple visual directions, or automate a generative model pipeline. Its Apps, node Workflows, third-party model access, and Enterprise controls are capabilities Valmera does not try to match.

Valmera is the stronger choice when the brief says remove weak takes, restructure a long recording, find exact moments, preserve factual source footage, add captions and overlays, reframe for delivery, mix sound, and return a complete rendered edit without operating a visual node graph or manual timeline.

Agent's Final Cut timeline makes Runway more capable than a pure generator and can assemble uploaded footage, generated clips, voiceover, and music. Valmera remains more directly designed around handing an open-ended editorial brief to an agent. A buyer should test the same real project rather than compare both products by the generic phrase AI video editor.

The products can complement each other. Generate or transform a difficult insert in Runway, export it, and place it inside a broader Valmera or NLE edit. Keep source footage, approved masters, prompts, rights records, and deliverables outside either vendor so the production remains portable.

Runway's new Unlimited Mode changes the price comparison but not the category difference. For supported Pro or Max models, repeated generation can avoid credit charges while taking longer and using fewer concurrent slots. Valmera's credit meter pays for an agent turn that analyzes or edits uploaded footage. Compare approved deliverables, operator time, queue delay, source duration, and revision count—not the word unlimited against a raw credit balance.

Runway Studio and Valmera also expose control differently. Studio gives a visible multi-track surface for manual splitting, trimming, shot detection, track movement, and audio separation, while Agent can assemble around it. Valmera delegates editorial decisions through conversation and renders a versioned cut from real footage. A creator who wants to touch clips manually may prefer Studio; a creator who wants to describe an editorial result may prefer Valmera.

Seedance 2.5, Aleph 2.0, Kling, Veo, and Runway's other models solve generation or semantic shot transformation. Valmera does not need to imitate that catalog to be useful. A practical stack can generate a missing shot in Runway, export it with its prompt and rights record, then let Valmera or a conventional NLE place it within a longer source-faithful program.

For the full feature-by-feature comparison, read Valmera vs Runway.

Primary Sources Checked

• Runway — Creative pricing and model credit matrix
• Runway — Agent product overview
• Runway Help — Creating with Runway Agent and Final Cut
• Runway Help — Custom Agent generation preferences
• Runway Help — Agent Skills and custom skills
• Runway Help — Building and running Workflows with Agent
• Runway Help — Using Brand Kits with Agent
• Runway — Aleph 2.0 product overview
• Runway Help — Creating with Edit Studio
• Runway Help — Aleph 2.0 prompting guide
• Runway Help — ProRes and PNG sequence exports
• Runway — Workflows product overview
• Runway Help — Introduction to Workflows
• Runway Help — Creating with Apps
• Runway Help — Creating with Gen-4.5
• Runway Research — Introducing Gen-4.5
• Runway Help — Performance Capture with Act-Two
• Runway — Official MCP server
• Runway Help — How credits work
• Runway Help — Which plan is right for me?
• Runway Help — Free plan details
• Runway Help — Standard plan details
• Runway Help — Max plan details
• Runway Help — Unlimited is switching to Max
• Runway Help — Credit rollover and purchased credits
• Runway Help — Completed-generation credit refunds
• Runway Help — Workspace members and roles
• Runway Help — Managing a Workspace subscription
• Runway Help — Managing a subscription
• Runway Help — Requesting a payment refund
• Runway Help — Canceling a plan
• Runway Help — Charges after cancellation
• Runway Help — Supported file types
• Runway Help — Retired editor and replacement workflows
• Runway Help — Sharing and privatizing an asset
• Runway Help — Security and privacy standards
• Runway — Data Security
• Runway — Enterprise product and no-training promise
• Runway Help — When to choose Enterprise
• Runway Help — Enterprise features
• Runway Help — Enterprise third-party models
• Runway — Terms of Use
• Runway — Privacy Policy
• Runway Help — Voice verification and consent
• Runway Help — Creating a custom voice
• Runway Help — API FAQs and separate credits
• Runway Help — Linking Developer and Web App accounts
• Runway Help — Credit-spend notifications
• Runway Help — Generating in Unlimited Mode on Pro and Max
• Runway Help — Creating with Seedance 2.5
• Runway Help — Trimming and assembling clips in Studio
• Runway Help — Generating with Sessions
• Runway Help — Managing assets
• Runway Help — Creating with Kling 3.0 Motion Control
• Runway Help — Third-party video model documentation
• Runway — Public service status

Frequently Asked Questions

Want the AI to Perform the Edit?

Upload your footage to Valmera, describe the outcome, and refine the rendered edit through conversation.