How to Make AI Videos with Runway Gen-4: The Complete Creator’s Guide

Runway Gen-4 turns a still image into a 5- or 10-second video clip. You supply the first frame and a short prompt describing the motion, and Gen-4 animates it — a person walks, a camera pushes in, rain starts to fall. A 5-second clip costs 60 credits on full Gen-4 or 25 credits on the cheaper, faster Gen-4 Turbo, and the output is 720p-class video at 24 frames per second. Gen-4 makes no sound at all, so dialogue and music have to be added afterwards. This guide walks through the whole workflow — setup, first clip, prompting, consistent characters, Act-Two performance capture, audio, editing, and pricing — using Runway’s own documentation as read on 9 November 2026. It is a documentation-based guide, not a hands-on test: every fact here traces to a Runway help-centre article or Runway’s pricing page, listed in the Sources section at the end.

What Runway Gen-4 Is — and Where It Sits in Late 2026

<img src="

” alt=”A still photograph print beside a small video camera, echoed as a moving frame on a monitor behind it”>

Gen-4 is Runway’s family of image and video generation models, and it is best understood as a shot maker, not a scene maker. Its core skill is image-to-video: you give it a starting image, and it produces a short clip — five or ten seconds — in which that image comes alive. The image becomes the first frame of the clip, and your text prompt describes what should move. Gen-4 Turbo is the faster, cheaper version of the same idea. Gen-4 Image is the still-image generator that produces the starting frames. And Gen-4 References is the system that keeps the same character, object, or style consistent from one shot to the next.

One honest framing matters before anything else. Gen-4 is not Runway’s current flagship model. That title belongs to Gen-4.5, which Runway calls “state-of-the-art” and the standard for motion quality, prompt adherence, and visual fidelity. Gen-4.5 adds text-to-video and flexible durations of two to ten seconds, and Runway’s prompting guide is now optimised for it. Gen-4 remains available, though, and it is the model this guide teaches: it is the workhorse for image-to-video shots, it is cheaper to experiment with through Turbo, and everything in this guide — the first clip, the prompting method, References, Act-Two, the editing tools — applies to the way most creators actually use Runway day to day. Where Gen-4.5 changes the picture, this guide says so explicitly.

That framing also sets the right expectations. Gen-4 produces clips, not films. Each generation is five or ten seconds long, at 720p-class resolution, with no native audio. A finished video is therefore something you assemble: a sequence of short shots, planned in advance, stitched together in an editor, with sound layered on afterwards. Creators who accept that constraint and plan around it get far more from Gen-4 than creators who expect it to produce a finished thirty-second scene in one pass.

Runway is one of several serious tools in this space, and whether it is the right one for a given project depends on what matters most — native audio, clip length, control, or cost. If you are still choosing between tools rather than learning one, our five-way comparison of AI video generators puts Runway alongside Sora, Veo, Kling, and Pika, with side-by-side tables and a choosing workflow. This guide assumes you have picked Runway and want to use it well. For the wider story of how the field got here, see also the AI video race after Sora, which explains why the current generation of tools looks the way it does.

Setting Up: Your Account, Free Credits and the Workspace

Getting started with Runway takes minutes. You create an account on Runway’s site, and the Free tier gives you 125 one-time credits that do not expire, plus 5 GB of asset storage. Runway describes the free tier as “a selection of generative creative models to test concepts” — it does not name which models are included, so the honest advice is to open the model picker in the app and check what your plan unlocks. Gen-4.5 and Act-Two both require a Standard plan or higher, and that requirement is stated explicitly in their help articles. Gen-4’s own article states no plan requirement, so do not assume from the pricing page alone whether full Gen-4 is available on Free; check the picker.

Most of the work in this guide happens in Tool mode, the part of the Runway web app organised by what you want to make — Video, Image, and so on — rather than in Apps, where standalone tools such as the Video Backdrop app live. Completed generations save to Assets, with finished videos filed under Assets and then Sessions by default, so your outputs are findable even if you generate a dozen variants of the same shot in one sitting.

What each plan unlocks matters because several features in this guide are gated. The four plans, as read on Runway’s pricing page on 9 November 2026, are: Free at $0 with 125 one-time credits; Standard at $15 per month, or $12 per month billed annually, with 625 credits a month; Pro at $35 per month, or $28 billed annually, with 2,250 credits; and Max at $95 per month, or $76 billed annually, with 9,500 credits. Standard unlocks all AI video and image models, music, voice and SFX generation, five parallel generations, up to three projects, no watermarks, 4K upscaling, and 20 GB of storage. Pro adds fifteen parallel generations, up to five projects, a Brand Kit, a custom voice clone, top-up credits at any time, Runway MCP, unlimited 4K upscaling, and 100 GB of storage. Max adds twenty parallel generations, up to ten projects, up to three Brand Kits and three custom voice clones, HDR, ProRes and image-sequence export, and 500 GB of storage. The full pricing picture, including worked cost examples, is in the pricing section near the end of this guide.

One practical point before you generate anything: Runway’s generation model is credit-based, and every model charges a set number of credits per generation, by model, duration, and resolution. Credits are the unit you budget. The discipline that runs through this guide — draft cheap in Turbo, spend full Gen-4 credits only on shots that already work — exists because credits are finite, and a habit of generating everything at full price burns through a monthly allowance in an afternoon.

Your First Clip: Image-to-Video Step by Step

Your first Gen-4 clip follows a short, fixed sequence. Open Tool mode, go to the Video tab, and choose your model from the dropdown — Gen-4 or Gen-4 Turbo. Upload your input image: this step is not optional. Gen-4 is image-to-video on both Gen-4 and Gen-4 Turbo, so a starting image is required, and it becomes the first frame of your clip. Then write your prompt. The prompt’s job is to describe motion, not to describe what the image already shows — the image already carries the composition, the subject, the lighting, and the style. A prompt of up to 1,000 characters is allowed, but short and specific beats long and vague.

Next, set the duration: five or ten seconds, and nothing in between. Runway’s own guidance is that simple actions suit five seconds, while prompts with multiple movements need ten, so the model has time to complete them naturally. Choose one of the six resolutions — 16:9 at 1280×720, 9:16 at 720×1280, 1:1 at 960×960, 4:3 at 1104×832, 3:4 at 832×1104, or 21:9 at 1584×672. The fixed-seed setting is optional and off by default; reusing a previous seed gives generations with similar motion and style, which is useful when you are iterating on a shot rather than starting over. Then generate.

The cost of that first clip depends on the model and the length. A five-second clip costs 60 credits on full Gen-4 (12 credits per second) or 25 credits on Gen-4 Turbo (5 credits per second). A ten-second clip costs 120 credits on Gen-4 or 50 on Turbo. That is why Runway itself recommends testing in Turbo first: it is cheaper and faster, and you should switch to full Gen-4 only when a shot needs it. A first clip is exactly the kind of experiment Turbo exists for.

Once the clip is generated, the controls underneath it let you favourite it, upscale it, or download it. Completed videos save to Assets, under Sessions, by default. From a finished clip you also get a set of “Use” options that open the next steps of the workflow: Use as Character or Use as Performance in Act-Two, Retime Video, Expand Video (which opens Edit Studio), Use Current Frame, and Upscale to 4K. Each of those is covered in its own section below. For now, the habit to build is simple: generate, watch the whole clip, and only then decide whether the shot works, needs a prompt tweak, or needs a different starting image. Most first clips need at least one retry, and that is normal — the iteration loop is the workflow.

The Starting Image: Picking or Generating a First Frame

Because the input image becomes the first frame, the starting image is the single most important creative decision in a Gen-4 shot. It sets the composition, the subject, the lighting, and the style, and the model will not fix a weak image for you — it will animate it. A blurry input gets worse in motion, not better: Runway’s prompting guide warns that blurry hands and faces in the input image get intensified in video. Start with a high-quality, artefact-free image, and treat the first frame as a photograph you are proud of, not a sketch you hope the model will rescue.

You have two sources for that image: upload your own, or generate one with Gen-4 Image. Uploading your own is the right choice when the shot depends on a real place, a real person, or a specific photograph. Generating with Gen-4 Image is the right choice when you need a controlled starting frame — a particular composition, a consistent character, a scene that does not exist yet. Gen-4 Image costs 8 credits per 1080p image or 5 credits per 720p image, with prompts of up to 1,000 characters and the same six aspect ratios as video. You can generate a single image or a batch of four, set the Aesthetic Range from 0 to 5 — where 0 means no variation from the prompt and 5 means the most dramatic variation — and fix a seed when you want reproducible results.

Gen-4 Image has two features that matter for this workflow. Describe Image auto-generates a prompt from an uploaded image, which is useful when you have a reference photograph and want to know how Runway reads it. Vary generates a new image set in the style of a selected output, with different structure and composition — a quick way to explore variations of a frame you almost like. And the Use option loads the image straight into the generative video tools, with Text/Image to Video as the default — the still-to-clip handoff, in one click, with no downloading and re-uploading in between.

One trap deserves attention before you animate anything: input images carry implied motion cues. A photograph with motion blur, a subject caught mid-action, a dust cloud, or directional lines already suggests movement, and if your prompt asks for motion that contradicts those cues, the model has to fight the image. The result is usually more iteration, not a better clip. Runway’s advice is to remove or minimise the cues first — for example with Text/Image to Image — and then animate the cleaned frame. A still, neutral starting image is easier to direct than a busy one, and the five seconds of motion you add will read more clearly against a calm frame.

Writing Motion Prompts That Actually Work

Prompting for Gen-4 is a different skill from prompting for image generation, and the difference is the whole point. In an image prompt you describe what things look like. In a Gen-4 prompt you describe what should happen. The image already defines the composition, the subject, the lighting, and the style, so re-describing those wastes characters and can confuse the model. The prompt’s job is motion, and motion has five components: the subject’s action, environmental motion, camera motion, motion style and timing, and direction and speed. You do not need all five in every prompt — omitting components gives the model creative freedom — but knowing the five is what lets you diagnose a clip that came out wrong.

Runway suggests a simple structure: “The camera [motion description] as the subject [action]. [Additional descriptions].” Structure and word order matter less than clarity and low ambiguity, but the pattern is a good default: camera first, subject second, extras last. For example, a prompt in this shape might read: “The camera slowly pushes in as the subject walks forward through the market. Dust drifts in the warm afternoon light.” Every element in that sentence is a motion instruction, not a visual description. If you are new to this, start simple and add detail to refine — a short prompt that works is a better foundation than a long prompt that fights itself.

Refer to the people and objects in the frame with general language — “the subject”, “the woman” — rather than re-describing what they look like. General references isolate the thing that should move without inviting the model to redraw it. Describe visuals only when you have a reason to: when you are introducing an element that is not in the image, when something changes dramatically, when a transformation needs detail, or when two or more elements interact. If you are learning the mechanics behind this, our plain-language explanation of how text-to-video generation works covers the underlying idea of turning a prompt into moving pictures.

For shots with more than one beat, Gen-4 supports sequential prompting. You can write it in natural language — “X occurs, then Y occurs. Finally, Z occurs.” — or with timestamps, such as “[00:01] X occurs. [00:03] Y occurs.” Match the complexity of the sequence to the duration: a five-second clip has room for about one clear action, and a ten-second clip can hold two or three beats if they are simple. Trying to cram a three-act story into five seconds is the most common prompting mistake, and the fix is almost always a longer duration or a simpler prompt, not more words.

Longer sequences are built by chaining clips, and the technique is worth learning early. Take a finished clip, scrub to its end, and use the Use Current Frame option to extract the last frame; that frame becomes the input image for the next clip. Join the clips in a video editor afterwards, removing the shared frame so the cut is invisible. The same idea works with timestamps inside a single prompt, but chaining is how you get past the ten-second ceiling, and it is covered in full in the stitching section later in this guide.

When a clip comes out wrong, Runway’s troubleshooting guidance covers the three failures you will meet most. Unwanted cuts — the shot jumping when it should flow — usually respond to a longer duration and a prompt component such as “Continuous, seamless shot”. Minimal motion, where the clip barely moves, responds to describing what motion should occur explicitly, or to a line such as “The locked-off camera remains perfectly still” or “The camera must start and end on the exact same frame to create a perfect loop” or “Minimal subject motion only”, followed by stabilisation in an editor. And when the input image itself is the problem, the answer is to fix the image before you animate it — a still frame with contradictory motion cues or visible artefacts will keep producing the same failure no matter how you reword the prompt.

Gen-4 vs Gen-4 Turbo vs Gen-4.5: Which Model for Which Shot

Three models share the Gen-4 name in different ways, and picking the right one for each shot is the cheapest skill in this guide. The table below is built from Runway’s own help articles, as read on 9 November 2026.

Gen-4 and Gen-4 Turbo are the same kind of tool at different prices. Both take an image and animate it for five or ten seconds at 24 frames per second, in any of the six aspect ratios. Turbo costs less than half as much per second — 25 credits for five seconds against 60 — and it is faster. The trade-off is quality: full Gen-4 is the higher-fidelity model, and it is the one to use when a shot has already proved itself in Turbo and deserves the best rendering you can give it. Runway’s own recommendation is exactly that workflow: test in Turbo first, switch to full Gen-4 only when a shot needs it. Treating Turbo as the draft model and Gen-4 as the finishing model is the single most effective way to keep a monthly credit allowance alive.

Gen-4.5 is a different step up. It is the model Runway calls state-of-the-art, and it changes two things that matter for planning. First, it accepts text-to-video as well as image-to-video — so a shot can start from a written description instead of a still image — though text-to-video is 16:9 only, while image-to-video supports all six aspect ratios. Second, durations are flexible from two to ten seconds rather than fixed at five or ten, which makes it easier to match a clip to the action it contains. It costs the same as full Gen-4 at 12 credits per second, requires a Standard plan or higher, runs on the web only, and offers infinite generations in Explore Mode. Its prompting is the same motion-first style: image-to-video prompts focus on motion, while text-to-video prompts describe the visuals and the motion together.

The practical rule is simple. Draft and iterate in Turbo. Promote a proven shot to Gen-4 when fidelity matters. Reach for Gen-4.5 when you need text-to-video, a two- or three-second clip, or the best motion quality Runway currently offers — and remember that at the same 12 credits per second, Gen-4.5’s flexibility is what earns the price, not a cheaper rate. One cost note that applies to all three models: on Max, Unlimited (Legacy), and Enterprise plans, ProRes or PNG-sequence export can be selected at generation time for an extra 5 credits per second. For most creators that export option only matters once clips leave Runway for a professional editor, which is covered in the stitching section.

References: Keeping the Same Character, Object or Style Across Shots

A single good clip is easy. Ten clips featuring the same person, in the same style, in different places, is the hard problem — and it is the problem Gen-4 References exists to solve. References take one or more images and create new images using their characteristics, styles, characters, or objects. Runway says they “excel at generating consistent characters across different lighting conditions, locations, and treatments, all from just a single reference image.” The system works in the Image tab of Tool mode: toggle References on, drag in an image or pick one from Assets, and the reference is active for the next generation.

You can have up to three active References per generation, and you call a saved reference in your prompt with its name, using the @ symbol — for example, “@bryan wearing a denim shirt…” tells the model to use the saved reference called bryan and put him in a denim shirt. The reference holds the identity; the prompt describes everything that changes — the outfit, the location, the lighting, the action. That separation is the whole technique: the reference is what stays the same, the prompt is what changes.

Saving a reference properly is worth doing the first time. Uploads are temporarily saved in the active session only by default, so a reference you do not save disappears when the session ends. To keep one, hover over it, choose tag to save, give it a name, and press Enter — it then persists across sessions. You can delete a saved reference through Edit (the pen icon) and Remove (the trash icon), and you can share one with your workspace members through Edit and then Share with Workspace, so a team working on the same character uses the same reference instead of three near-identical ones.

Single-reference and multi-reference prompting are different tools for different jobs. With one reference, the prompt changes everything around the subject — a useful pattern is to describe the subject’s shoes or pants explicitly, because Runway notes that this helps consistently produce a full-body shot. With two or three references, you can combine a character, a location, and an object for precise control, and the results are more predictable. One practical tip from Runway: when you use an image that already contains a subject, cover the existing face with a black box in a photo editor before uploading, so the model is not confused about which face belongs to which subject.

References also work for places, not just people. A reference image can generate consistent environments by prompting for different angles, focal points, and objects within it — a cohesive setting with the same atmosphere across shots, which is how you build believable b-roll for a location that only exists as a generated image. And for composition control, there are sketch references: draw a new image or annotate an existing one to control composition, character gaze, or object placement. You can sketch a scene from an empty canvas, or sketch over a loaded reference using the marker icon; exporting the sketch loads it as a Reference. One limitation to know: sketches cannot be edited once exported, so get the sketch right before you export it.

The bridge from stills to motion is one click. Hover over any output image and the camera icon loads it into the latest video model — Gen-4 or Gen-4.5 — to animate it. That is the moment a reference workflow becomes a video workflow: the consistent character you built in stills becomes the first frame of a clip, and the motion prompt takes over from there.

Building a Consistent Character and Scene, Shot by Shot

References are the tool; the character workflow is how you use it across a whole project. The sequence that works is to lock the character in stills first, before any video is generated. Start with a good reference photograph. Runway’s recommendations are specific: natural, even lighting; moderate quality; a neutral expression. The neutral expression matters because it is a blank canvas — a face already mid-expression is harder for the model to transform into other expressions later. Stylised inputs can still work, but the plain, well-lit reference is the reliable starting point, and the photographer in the image above is setting up exactly that kind of shot.

From there, iterate one element at a time. Change the outfit, or the location, or the lighting — never all three at once — and use each result as a new reference by hovering over the output and choosing “Reference for image”. Changing one element at a time is what makes a failure diagnosable: if the character’s face drifts, you know which change caused it, and you can step back one version instead of starting over. Keep the character iteration and the scene iteration on separate paths, and combine them only once each is working on its own. That separation is slower in the first hour and much faster across a whole project, because a broken combination is otherwise impossible to debug.

Two of Runway’s practical tips are worth building into the habit. First, describe the subject’s shoes or pants in the prompt when you want a full-body shot — it anchors the framing and keeps the model from cropping to a head-and-shoulders portrait. Second, when a reference image already contains a subject whose face might confuse the model, cover the existing face with a black box in a photo editor before uploading. Both are small moves that prevent the most common consistency failures: a character who changes height between shots, or a face that quietly becomes someone else’s.

Once a still is approved, animate it. Hover over the output image, use the camera icon to load it into Gen-4, Gen-4 Turbo, or Gen-4.5, and write a motion prompt for it. The still you approved is the first frame, so the character, the lighting, and the composition are already locked — the only new variable is the motion, which is exactly the variable a motion prompt is built to control. The same logic applies to scenes and b-roll: a single reference image for a location, prompted for different angles, focal points, and objects, gives you a set of shots that feel like the same place, with the same atmosphere, even though each one was generated separately. A market, a forest, and a rooftop at dusk can all belong to one character’s world if the reference discipline holds.

Making Characters Talk and Move: Act-Two Performance Capture

Act-Two is Runway’s answer to a different problem from Gen-4’s: not “make this still move” but “make this character perform”. It animates a character using a driving performance video — you film someone acting out a scene, supply a character reference as an image or a video, and Act-Two transfers the movement, the speech, and the expression from the performance onto the character. It is not a text prompt. The driving performance captures the movement, the expressions, the audio, and the gestures, and the character inherits them.

Act-Two requires a Standard plan or higher, runs on the web only, and costs 5 credits per second with a three-second minimum — a performance under three seconds costs 15 credits — up to a maximum of thirty seconds. The output uses the same six resolutions as Gen-4, at 24 frames per second. The choice of character input changes what Act-Two controls. With a character image, you get automatic environmental motion and gesture control: poses, gestures, and body motion transfer from the performance. With a character video, Act-Two keeps the original video’s subject, environment, and camera motion, and controls facial movement and expressions only — there is no gesture control on video inputs. That distinction decides which input to use: an image when you want the character to move through a performance, a video when you only want the face to change.

Two settings shape the result. Gestures can be switched on or off, and only applies to image inputs; with it off, you get facial and environmental motion only. Facial Expressiveness defaults to 3, and Runway advises testing at the default first: lower values are less expressive but may improve consistency, while higher values are more expressive but may cause artefacts. As with Gen-4, the default is the right starting point, and the extremes are for fixing specific problems.

Filming a good driving performance is a craft in itself, and Runway’s best practices are worth following literally. One subject only. The face visible throughout. Frame waist-up at the furthest. Well-lit, with defined features. No cuts. Hands in frame at the start, so gestures have something to transfer from. Start in a similar pose to the character input, so the transfer has a clean starting point. Move naturally, not excessively or abruptly — and know that some expressions, such as sticking out a tongue, are not supported. A performance that follows these rules transfers cleanly; a performance that breaks them produces the artefacts that make Act-Two look unreliable when the tool is not the problem.

One edge case to plan around: if the performance is longer than the character video, the character video loops with a reversed “boomerang” effect. Prefer a character image, or a character video of matching length, for natural results. After generation, the Use menu offers Change Voice, which swaps the voice — preview the voices, then generate, and note that the generated voice may match the accent of the original audio. Clear audio with minimal background noise and consistent volume and pitch gives the best results. Runway also documents a multi-character dialogue workflow in its help centre, for readers who need more than one character speaking; the steps are not covered here because that article was not part of this guide’s research.

A note on history, because old tutorials will mislead you: the standalone Lip Sync tool was deprecated on 27 May 2026, and Act-Two is its replacement. If a guide from before that date tells you to open Lip Sync, it is describing a tool that no longer exists. Route dialogue through Act-Two.

Sound: What Gen-4 Can’t Do — and How to Add Audio Anyway

The honest headline first: Gen-4 clips are silent. Gen-4 generates no native audio at all — Runway’s own resources page states that Gen-4.5 “doesn’t generate embedded audio itself”, and the Gen-4 help article lists no audio capability whatsoever. If you need a clip with sound, that sound has to come from somewhere else. This is the single most common misunderstanding about Runway’s video tools, and it is better to learn it from this paragraph than from a finished clip that arrives without a soundtrack.

Runway’s own audio workflow has five steps. First, generate the video. Second, add dialogue or performance through Act-Two, which transfers a filmed performance — including its audio — onto a character. Third, layer sound effects and ambient audio from Runway’s generative audio tools, which generate sound from a text description and time it to the on-screen action. Fourth, add music or a voiceover separately. Fifth, refine the result with Aleph 2.0. Adding audio in post gives you independent control of each layer — you can change the music without re-rendering the picture, or replace a line of dialogue without touching the sound effects. Native generation is faster, but it couples picture and sound together, so a bad line of dialogue means regenerating the whole clip.

For voiceover, a dedicated voice tool is the practical route for most creators, and ElevenLabs is the one we have covered in depth on this site — its voice tools fit naturally onto a silent Gen-4 clip in an editor. For the music bed, ElevenLabs instant soundtracks generate music for a finished video, which is the same job with a different tool. The point in both cases is that the voice and the music are separate decisions from the picture, made after the picture is finished, and that separation is a strength rather than a limitation.

There is one in-Runway alternative for readers who need sound in the same pass as the picture. Two models hosted on Runway generate audio natively: Kling 3.0 generates dialogue, ambient sound, and effects together with the visuals, and Seedance 2.5 likewise returns multi-shot video with sound built in. They are not Gen-4, and they are not covered in depth here, but if your project cannot tolerate a separate audio pass, they are the route inside Runway’s platform. For the same question asked outside Runway, our guide to making AI videos with Google Flow and Veo covers Veo’s native-audio workflow, which is the main alternative named in this section.

Dialogue deserves a final caution. Runway’s resources page quotes Google DeepMind’s assessment that “natural and consistent spoken audio, particularly for shorter speech segments, remains an area of active development.” In plain terms, dialogue is the current frontier: short, simple spoken lines are workable, and long or complex dialogue is where the tools still struggle. Test a short clip before committing to a dialogue-heavy sequence, and budget for retakes on the audio in the same way you budget for retakes on the picture.

Editing Finished Clips: Retime, Expand, Upscale and Aleph 2.0

A finished Gen-4 clip is rarely the final version. Runway’s editing tools start with the free fixes and end with Aleph 2.0, a genuine prompt-based editor — and the order matters, because the free fixes solve most problems without spending a credit.

Retime Video is the first stop. From a finished clip you can trim it, adjust its speed, add handheld shake, or reverse it, and retiming costs no credits at all. Trimming starts an export within the session, so the trimmed clip is a new asset you can use downstream. Before reaching for anything more complex, ask whether the problem is really timing: a clip that is two seconds too long, or that needs to feel handheld rather than locked off, is a retiming job, not a regeneration job. The Use Current Frame option sits alongside retiming in the same menu, and it is the chaining tool from the prompting section — extract the last frame of a clip and use it as the next clip’s starting image.

Expand Video opens Edit Studio to expand a video into a new aspect ratio, powered by Aleph 2.0 — useful when a clip shot in 16:9 needs to become a 9:16 vertical for a phone screen, without cropping away the subject. Upscale to 4K is a separate action on a finished clip: 4K upscaling is a Standard-plan feature, and it is unlimited on Pro and Max. One thing to be clear about: an upscaled 720p clip is not a native 4K generation. Upscaling sharpens and enlarges what Gen-4 produced; it does not add detail the model never rendered. Treat it as a delivery step for screens that need 4K, not as a quality upgrade for a weak shot.

Edit Studio, powered by Aleph 2.0, is the real editor. It edits existing footage — traditional or generated — with prompts: swap products, replace characters, transform shots, remove or replace backgrounds, remove unwanted objects, insert elements or effects, across single shots or multi-shot sequences of up to thirty seconds. The input limits are specific: a conventional aspect ratio only; longer than two seconds and shorter than thirty (anything over thirty seconds is auto-trimmed); 480p to 1080p, with 720p or 1080p recommended; 24 to 30 frames per second (anything over 30 is downsampled); and no more than ten cuts or shot changes — eleven or more is an error. If your clip fails to load, check those limits first.

Cost is per second of edited footage: Aleph 2.0 runs at 28 credits per second, with a 56-credit minimum. Keyframe iteration is billed separately by image model — Nano Banana Pro at 20 credits per image, GPT Image 2 at 5 credits per image, and Gen-4 Image at 8 credits per image. Ranged edits charge only for the edited portion, not the full clip, which makes a small fix to a long clip much cheaper than re-editing the whole thing.

The workflow, in Single edit mode — Multi-edit and Expand modes are listed as coming soon — is: upload a clip or pick one from Assets; select a keyframe that clearly shows the edit target, using the widest shot for scene or background changes and a close frame for small details; write a short prompt describing the change, optionally with a reference image; preview the keyframe edit first, using the Prompt versions tray to reuse a prompt, edit the image, or preview before and after; then generate the video. The top thumbnail is always the original, so clicking it starts fresh. Iterations stack, and each version can be built on further — so an edit that is nearly right is a starting point, not a dead end. Extra motion is optional: a text prompt for motion that would not happen on its own, such as “fire begins to spread up the trees”. It is usually unnecessary, because Aleph 2.0 animates adjustments from the keyframe edit alone.

Prompting Aleph 2.0 follows a simple formula: an action verb plus a description of the transformation. Runway’s own examples make the pattern clear — “Change the outfit and bag to soft yellow” — and the model changes only what is asked, preserving everything else, including the background, lighting, and surrounding details. The documented use cases cover most real editing jobs: change the background (“change background to ancient city ruins”); change wardrobe or accessories (“Change outfit to bright yellow fur coat”); change the weather, season, or time of day (“heavy overcast and raining”); relight a shot (“Chiaroscuro lighting”); replace an object or subject (“Swap for black leather clutch with gold buckle”); and restyle a whole shot (“1920s silent era film”). Simplicity, iteration, and precision are the three rules Runway gives for Aleph prompting, and they are the same three rules as for Gen-4 itself. One historical note: Aleph began as Runway’s in-context video editor in 2025, but the current tool is Aleph 2.0 inside Edit Studio — Runway’s help centre documents no standalone Gen-4 Aleph article today, so this guide presents Aleph 2.0 as the tool to learn.

Stitching Short Clips into a Finished Video

The five-to-ten-second ceiling is the defining constraint of Gen-4, and the answer to it is planning. A finished video is a sequence of shots, and each shot is a separate generation with its own starting image and motion prompt. Plan the sequence before you generate anything: a shot list with a first frame and a motion description for each clip. Without a shot list, you generate a pile of disconnected clips and discover in the editor that nothing matches. With one, the clips arrive already belonging together.

The mechanical technique for continuity is last-frame chaining. Finish a clip, scrub to its end, use the Use Current Frame option to extract the last frame, and use that frame as the input image for the next clip. Join the clips in a video editor afterwards, removing the shared frame so the cut is invisible. The chain is only as strong as the discipline behind it: keep the same references active across shots, reuse a fixed seed where it helps, and keep the same style language in the prompts, so the lighting and the character stay consistent from one clip to the next. A clip that changes the character’s jacket halfway through a sequence is a continuity error, and it is cheaper to prevent with references than to fix in the editor.

A realistic short project shows how the numbers add up. Take a thirty-second product spot built from six five-second shots: an opening wide shot of the product in its setting, a close-up of a hand reaching for it, a detail shot of the product in use, a lifestyle shot of the product in a room, a second angle for variety, and a closing shot with the product centred. Each shot is a separate Gen-4 generation with its own starting image — six stills, six motion prompts, six clips. At Turbo rates that is 150 credits for the picture; at full Gen-4 rates it is 360. Each shot then gets retimed or trimmed for free if needed, and the six clips are joined in an editor with sound layered on top. Nothing in that recipe requires a tool beyond what this guide has already covered — it is the same mechanics, repeated six times, with a shot list holding it together.

For projects heading into a professional editor or a visual-effects workflow, export quality matters. On Max, Unlimited (Legacy), and Enterprise plans, ProRes or PNG-sequence export can be selected at generation time for an extra 5 credits per second. The same export option applies to Edit Studio outputs. For most creators the standard download is enough; ProRes and image sequences earn their cost when the footage will be graded, composited, or cut against camera-original material, where a compressed delivery file would fall apart.

Pricing, Credits and What a Finished Minute Really Costs

Runway’s pricing is credit-based, and the honest way to understand it is to work from the credit rates rather than the plan names. The table below is taken from Runway’s pricing page as read on 9 November 2026. All prices are in US dollars.

The credit rules in the pricing FAQ decide how far those allowances actually go. Credits are the unit spent to generate images, video, and audio, and each model uses a set number per generation by model, duration, and resolution. Standard and Pro credits do not roll over — they reset within 24 hours of the billing date — so an unused balance at the end of the month is lost. Max credits roll over for up to one month. Purchased top-up credits never expire, and the minimum top-up purchase is 1,000 credits. No Team or Enterprise prices appear on the pricing page, so this guide does not quote any.

Worked arithmetic, on the stated rates, makes the plans concrete. A five-second Gen-4 clip costs 60 credits, so Standard’s 625 credits buy about ten such clips a month if nothing else is spent. The same five seconds in Turbo costs 25 credits, so Standard buys about twenty-five Turbo clips. A ten-second Gen-4 clip costs 120 credits, so Standard buys about five. On Pro’s 2,250 credits, the same maths gives about thirty-seven five-second Gen-4 clips, or ninety Turbo clips. On Max’s 9,500 credits, it gives about 158 five-second Gen-4 clips, or 380 Turbo clips. These are arithmetic on the published rates, not promises — every image generation, Act-Two performance, and Aleph edit draws from the same balance, so a real month’s output is always less than the headline figure. One quirk worth knowing: Runway’s own pricing-page equivalents imply a slightly higher Turbo rate than the help article’s 5 credits per second — the pricing page’s 104 seconds of Turbo on Standard works out to about 6 credits per second. The two Runway pages disagree with each other, so treat the pricing page’s equivalents as headline figures and the per-second rates as the billing basis.

The number that matters most for budgeting is not the cost of one clip but the retake multiplier. Almost no shot is a keeper on the first generation. Budget several generations per keeper — a draft in Turbo, a refinement, then a final at full quality — and the cost of a finished minute of video becomes several times the cost of a minute of raw generations. That is why the Turbo-first discipline from the model section is the money-saving method in this guide rather than a minor tip: drafting at 5 credits a second instead of 12 is what keeps a thirty-second project inside a Standard plan instead of blowing through it. For a deeper look at how video tools compare on cost across the field, see what AI video tools cost and what they do.

Two pricing details deserve care. First, watermarks: the pricing page lists “No watermarks” as a Standard feature, but it does not explicitly state that Free outputs are watermarked. Paid plans advertise no watermarks; do not assume more than that about the free tier. Second, no promotional price appears in this guide — a time-sensitive banner on the pricing page is marketing, not a plan fact, and it is deliberately left out.

Limits, Honest Expectations and When to Use Another Tool

Gen-4 is a strong tool inside its limits and a frustrating one outside them, so it is worth naming the limits plainly. The hard ceilings are fixed: five or ten seconds per clip, 720p-class output, no native audio, and Standard-or-higher gates on Gen-4.5 and Act-Two. None of those change with better prompting. A project that needs a minute-long continuous take, or dialogue generated in the same pass as the picture, is a project for a different tool, not a better Gen-4 prompt.

The failure modes are mostly diagnosable, and Runway’s own documentation implies most of them. Bad inputs produce worse outputs: blurry hands, faces, or text in the starting image get intensified in video, so the fix is a better first frame, not a better prompt. Implied motion cues in the image — blur, dust, a mid-action pose — fight prompts that ask for different motion, so clean the cues first. Unwanted cuts respond to a longer duration and a “Continuous, seamless shot” prompt component. Minimal motion responds to explicit motion description or a loop instruction, followed by stabilisation in an editor. And dialogue is still the frontier across the whole field: test short before committing to a dialogue-heavy sequence.

A short note on retired tools, because old tutorials and search results will send you looking for them. Gen-3 Alpha was retired on 8 July 2026, and Gen-1 and Gen-2 are deprecated — do not follow any Gen-3 workflow as though it were current. The standalone Lip Sync tool was deprecated on 27 May 2026 and replaced by Act-Two. Remove Background was deprecated on 10 April 2026, with its function moving to the Video Backdrop app. Super-Slow Motion was deprecated, with its function moving to the Edit Video app. And the All Tools page itself was removed on 10 April 2026, with the standalone “Magic Tools” moving into Apps and Tool Mode. If a search result points you at a help page for one of these, read the page — it will tell you the feature is no longer available — and use the replacement named here.

When should you reach for another tool instead? Two exit ramps cover most cases. If the project needs native-audio dialogue generated in the same pass as the picture, Veo through Google Flow is the route to learn — our guide to making AI videos with Google Flow and Veo covers that workflow. If the project needs long clips rather than short ones, Kling is the tool most often named for the job, and our comparison of the five main AI video generators is the place to weigh that choice. And if you want several models under one roof rather than one subscription per tool, Krea puts Veo, Kling, and Runway in one browser studio. None of that is a verdict against Runway — it is the recognition that a shot maker is the right tool for shots, and that knowing when to stop using it is part of using it well.

Gen-4 rewards the creator who plans in shots, directs with motion-first prompts, keeps characters consistent with References, and budgets credits with Turbo-first discipline. Start with one clip. Make it in Turbo. Watch the whole five seconds. Then make the next one better.

Sources

All facts in this guide were checked against Runway’s own pages on 9 November 2026.

• Creating with Gen-4 Video — Runway Help Center
• Creating with Gen-4 Image References — Runway Help Center
• Creating with Gen-4 Image — Runway Help Center
• Creating with Gen-4.5 — Runway Help Center
• Image to Video Prompting Guide — Runway Help Center
• Creating with Edit Studio — Runway Help Center
• Aleph 2.0 Prompting Guide — Runway Help Center
• Performance Capture with Act-Two — Runway Help Center
• Runway pricing page
• Can AI generate videos with sound? — Runway Resources
• Deprecated Standalone Tools — Runway Help Center