To use Runway for a short AI video draft, begin with a rights-cleared one-shot brief and choose the mode deliberately. In image-to-video, the uploaded image acts as the first frame and sets composition, subject, lighting, and style, so the prompt should describe motion, camera work, and timing; in text-to-video, the prompt describes both look and motion.［2］［5］ Generate the smallest useful draft, review it frame by frame, and change one variable per iteration. A visually appealing draft is not automatically accurate, authorized, or ready to publish.

This guide creates a synthetic five-second concept: a paper boat crossing a shallow studio puddle while the camera slowly tracks beside it. It does not cover scriptwriting, storyboards, multi-shot continuity, distribution, or access troubleshooting. Those belong in the AI video overview, the script and storyboard workflow, and the Runway access guide.

Use the broader How to Use AI framework to define the human owner, evidence boundary, and final acceptance decision.

Key Takeaways

Clear rights, identity, and disclosure questions before uploading an input.

Use one shot, one subject action, and one camera behavior for the first draft.

Choose text-to-video for open visual invention and image-to-video when an approved image must anchor composition.

In image-to-video, prompt mainly for motion because the image already defines appearance.

Change one variable per iteration and record model, settings, date, input, and result.

Review continuity, anatomy, physics, text, identity, and safety frame by frame.

Keep publication approval separate from generation success.

What should the brief contain when you use Runway for a short AI video draft?

The first brief should be narrow enough that a reviewer can tell what succeeded. Runway's current generative-video guidance recommends starting with a clear idea and iterating rather than trying to force every detail into one generation.［1］

Purpose: internal motion concept, not final advertising
Duration: shortest available useful trial
Shot: one continuous side-tracking shot
Subject: unbranded paper boat
Action: boat drifts across a shallow studio puddle
Environment: controlled tabletop set, soft daylight
Camera: slow lateral track, stable horizon
Style: realistic editorial photography
Avoid: people, logos, readable text, additional boats, scene cuts
Input rights: synthetic brief; no third-party image uploaded
Acceptance: clear boat silhouette, plausible water, stable camera, no new objects
Owner: named human reviewer

Do not start with “make a cinematic product video.” That leaves subject, action, camera, continuity, rights, and success undefined.

How do you clear source and identity risks?

Before uploading an image or video, record where it came from, who owns it, what license or permission applies, and whether it contains a recognizable person, artwork, location, logo, confidential product, or personal data. Platform access does not grant rights to third-party material.

Use synthetic or commissioned inputs where possible. Do not imitate a living person's identity, fabricate an endorsement, or upload private footage without authority. Runway's terms and usage rules remain part of the release review, and the user is responsible for lawful inputs and outputs.［6］

Input	Minimum evidence	Common stop condition
Text-only brief	Author and intended use	Request impersonates or deceives
Your photograph	Original file and consent	Another person's rights unclear
Licensed stock image	Source page and license	License excludes intended use
Client asset	Written scope and owner	Confidential or approval boundary unclear
Public web image	Not sufficient by itself	No permission or provenance

This is an editorial workflow, not legal advice. Escalate uncertain rights, publicity, trademark, or sector-specific requirements to qualified owners.

Should you choose text-to-video or image-to-video in Runway?

Runway documents both text-to-video and image-to-video in its current video models.［2］ Text-to-video creates appearance and motion from the prompt, so it suits open exploration. Image-to-video uses the image as the first visual condition, so it suits an approved composition that needs motion.［4］［5］

Mode	You provide	Prompt emphasis	Main risk
Text-to-video	Descriptive text	Subject, action, environment, camera	Visual identity may vary widely
Image-to-video	Approved image plus text	Subject motion, environmental motion, camera	Source defects and rights carry forward

For the paper-boat concept, text-to-video is enough because no exact layout or identity must be preserved. If a client-approved still defined the composition, image-to-video would be the better controlled starting point.

Do not switch modes after every disappointing result. First identify whether the failure comes from insufficient visual anchoring, conflicting motion, or an overloaded brief.

How do you write a motion-first Runway prompt?

Runway's prompting guidance emphasizes direct visual descriptions and recommends thinking about what should appear and move.［3］ For image-to-video, the source image already supplies composition, lighting, subject, and style; restating those details can conflict with the input.［5］

A compact Runway text-to-video prompt:

A single unbranded white paper boat drifts steadily across a shallow studio
puddle. Soft daylight reflects on the water. The camera tracks slowly beside
the boat at water level with a stable horizon. One continuous realistic shot.

An image-to-video motion prompt for an approved boat still:

The paper boat drifts forward while small ripples spread behind it. The camera
tracks slowly to the right at the same height. The horizon and lighting remain
stable in one continuous shot.

Which motion components belong in an image-to-video prompt?

Runway's image-to-video guide lists five motion components: subject action, environmental motion, camera motion, motion style and timing, and direction and speed. You do not need all five; start with the one or two that matter most, and add detail while iterating. Its suggested order is camera motion, then subject action, then extra detail, though clarity matters more than order.［5］

Describe visual elements only when you introduce something absent from the image, request a dramatic change or transformation, or define how two elements interact. Check the input too: blurry hands, faces, or other artifacts can be intensified once the image moves, and motion blur or a mid-action pose implies a direction that a contradicting prompt will fight.［5］

Use positive visible instructions. A long negative list can make the prompt harder to interpret; keep exclusions in the brief and remove ambiguity from the main description. If a control or model behaves differently, check the current official guide rather than relying on an old prompt formula.

How do you generate the smallest useful trial?

Select the currently documented model and mode that match the brief, but do not hardcode a model name into an evergreen workflow. Model availability, duration, resolution, controls, and interface labels can change. Record what you actually used and the check date. At the check date, Runway's Gen-4.5 article lists text-to-video and image-to-video, durations from 2 to 10 seconds, web availability, and a Standard-or-higher plan requirement; you open it from Apps by searching for Gen-4.5, or from the Tools model selector with Video selected.［2］

Generate one result at the shortest useful duration and a reviewable resolution. The purpose of trial one is to test the shot definition, not to produce a campaign asset. Save this run record:

Field	Record
Run ID/date	Local identifier and exact date
Product surface	Runway web or approved workspace
Model/mode	Exact current selection
Input provenance	Text owner or image source/rights
Prompt version	Immutable prompt text
Settings	Duration, aspect, seed or controls shown
Output status	Draft only
Review result	Accepted parts, defects, next variable

Do not infer settings that are not displayed. If the platform does not expose a seed or deterministic replay, record that limitation.

How should you review the first draft?

Watch at normal speed, slowly, and frame by frame. Inspect the beginning, middle, end, and transitions between them.

Review these dimensions:

Subject stability: Does the boat keep one shape, count, material, and scale?
Motion: Does it drift rather than teleport, melt, or reverse without cause?
Camera: Is the lateral track smooth, with a stable horizon and no unintended cut?
Environment: Do water, reflection, shadow, and contact behave plausibly?
New content: Did a logo, word, person, object, or unsafe element appear?
Intent: Does the clip communicate the brief without requiring an explanation?
Release risk: Could viewers mistake synthetic content for documentary evidence?

Reject a draft that creates a different boat, impossible geometry, unexplained extra objects, identity confusion, or misleading real-world claim. Compression cannot repair a semantic defect.

How do you iterate one variable at a time?

Name the largest defect and change one class of instruction. If the camera shakes, revise camera behavior while holding subject, action, environment, mode, and other settings constant. If motion is correct but the boat changes shape, strengthen subject continuity in the next prompt.

Iteration	Changed variable	Held constant	Result
V1	Baseline prompt	Brief and mode	Camera too fast
V2	“tracks slowly” reinforced	Subject, action, light	Speed improved; horizon drifts
V3	Stable horizon clarified	All other wording	Candidate for review

This is not a scientific benchmark unless the tool exposes deterministic controls and the test method is defined. It is an auditable creative comparison. Keep rejected results only according to workspace policy; do not label them as approved assets.

For still-image prompting principles that can inform an approved anchor frame, see better AI image prompts.

What separates a draft from a publishable asset?

Generation success only means the platform returned a file. Publication requires a different gate:

provenance and permission for every input;
review for recognizable identity, logos, art, locations, and private information;
frame-level visual and technical quality review;
factual and contextual review of any implied real event or product behavior;
accessibility, caption, audio, and localization decisions;
disclosure or labeling required by platform, organization, client, or law;
named human approval for the exact final export and intended channel.

Do not use an AI video draft as evidence that a product works, an event occurred, or a person endorsed a claim. If the draft shows a real brand or person unexpectedly, stop and escalate rather than hiding it with a casual crop.

How should you archive the creative record?

Keep the approved brief, source provenance, prompt versions, model/mode and settings, review notes, rejected reasons, selected draft, required disclosures, and final approval. Use stable IDs so another reviewer can match the decision to the exact export.

Record limitations honestly: model controls may change; outputs may vary; a recreated run may not match; visual review does not establish legal rights. The general AI workflow provides a reusable verification boundary.

Final checklist

The purpose, shot, subject, action, camera, environment, and exclusions are explicit.
Every uploaded input has provenance and permission for the intended use.
Text-to-video or image-to-video was chosen for a stated reason.
The first run uses one shot and the smallest useful trial.
Model, mode, settings, prompt, date, and output ID are recorded.
The clip was reviewed at normal speed, slowly, and frame by frame.
Each iteration changes one variable and records the result.
Draft status is separate from final publication approval.

FAQ

Should I begin with text-to-video or image-to-video?

Use text-to-video when the look is open to exploration. Use image-to-video when an approved, rights-cleared image must anchor composition and identity.［4］［5］

How long should the first draft be?

Use the shortest duration that demonstrates the intended motion and camera behavior. Runway currently lists 2 to 10 seconds for Gen-4.5 and notes that prompts with several sequential actions may need a longer duration.［2］ A small trial is easier to review and revise than an overloaded sequence.

Should an image-to-video prompt repeat everything in the image?

Usually focus on the desired motion and camera behavior because the image already defines appearance. Add visual detail only when it resolves a specific ambiguity.［5］

Can I upload an image found through web search?

Search visibility is not permission. Use an original, commissioned, or appropriately licensed source and record its provenance and intended-use rights.

How many things should I change between generations?

Change one meaningful variable whenever possible. Otherwise you cannot tell which revision caused an improvement or regression.

Does a clean output mean it is ready to publish?

No. Rights, identity, accuracy, disclosure, accessibility, localization, channel policy, and human approval remain separate checks.

What if the output contains a logo or recognizable person?

Stop and review rights, identity, and context. Do not imply endorsement or conceal the issue with an unapproved edit.

Can Runway guarantee physical realism or continuity?

No. Review every result for changing objects, anatomy, reflections, contact, text, camera motion, and continuity. Model quality does not eliminate output review.

Does a VPN establish rights or unlock a Runway feature?

No. A VPN does not create copyright or publicity rights, grant account eligibility, approve content, or improve a generated clip's quality.

Disclaimer: This article provides general information, not legal or other professional advice. Rules, rights, and platform features may change; consult qualified owners for your intended use.

Sources

Runway, “Getting Started with Generative Video” — https://help.runwayml.com/hc/en-us/articles/37425232841875-Getting-Started-with-Generative-Video
Runway, “Creating with Gen-4.5” — https://help.runwayml.com/hc/en-us/articles/46974685288467-Creating-with-Gen-4-5
Runway, “Introduction to Prompting” — https://help.runwayml.com/hc/en-us/articles/47313698911891-Introduction-to-Prompting
Runway, “Text-to-Video Prompting Guide” — https://help.runwayml.com/hc/en-us/articles/42460036199443-Text-to-Video-Prompting-Guide
Runway, “Image-to-Video Prompting Guide” — https://help.runwayml.com/hc/en-us/articles/48324313115155-Image-to-Video-Prompting-Guide
Runway, “Terms of Use” — https://runway.com/terms-of-use

Sources checked 4 October 2026.

Image to Video Tips: Use Runway for a Short AI Video Draft