Choose an AI Video Workflow: Runway, Veo and a Conditional Kling Check

Revised October 15, 2026: for an English-speaking creator or small team making product demonstrations or e-commerce clips. We replaced a broad roundup with a buying protocol based on readable documentation. We did not generate comparison clips or measure quality. Account access and client rights remain purchase conditions.

Define the deliverable before the model#

Write a brief with approved product images, script, aspect ratio, language, territory and channel. State what must stay exact: label, colour, spoken claim and authorised likeness. Attractive footage that changes those facts fails the task.

Investigate Runway for an app experiment with documented plans. Investigate the Veo developer route if you need an API and can operate it. Keep Kling conditional until its current model, checkout and rights terms can be read. These are scope-based starting points, not quality rankings.

Evidence and cost checked October 15#

• Runway: its pricing page lists Gen-4.5 and other models, Free with 125 one-time credits, and Standard at $15 monthly or $12/month billed annually. Free credits are not a monthly grant. Model use, failed attempts and repair work affect usable-clip cost. Confirm checkout currency, tax and cycle; the displayed dollar amount is not a universal offer.
• Veo: Google's January 13 announcement documents enhanced Veo 3.1 in the Gemini API. It establishes that developer surface, not live access for every account today. Current per-second pricing and free entitlement were not established here; obtain the exact endpoint quote.
• Kling: the official entry point was not readable in this round. Version, price, quota, countries and commercial terms remain unverified. Do not subscribe or upload client assets merely because it appears in a comparison.

Exclusions and rights conditions#

OpenAI's deprecation table lists the Videos API and named sora-2 aliases/snapshots as removed September 24, 2026. That scope is excluded. It does not establish the app's shutdown date or a ChatGPT migration; see the scoped Sora record.

Vidu, Hailuo, Pika and PixVerse are not declared inferior: current plan/licence evidence is outside this shortlist. Open weights require a separate exact-version licence, hardware and deployment review. The historical comparison correction withdraws unsupported duration, speed and price assertions.

A price card is not a commercial licence. Before a client trial, save dated output-use, music, likeness, upload, retention, training, watermark and redistribution terms. Obtain authority for the source assets. Where region/account access or permitted use cannot be established, mark the candidate unverified and use an approved alternative. Worldwide availability and every plan's rights have not been checked here.

A repeatable shot trial#

This is a proposed experiment. Use only synthetic or authorised material: the same product still, short English script and three briefs — label close-up, camera move around the object, and a presenter only if likeness permission exists.

• Record date, vendor, app/API, model/version, plan, resolution, duration and settings. Use the same feasible duration and assets; record unavailable settings.
• Save every prompt, visible seed, output and rejected attempt. Set a spend/time cap first; do not count only successful clips.
• Check product/text fidelity, continuity, unwanted changes, audio sync where applicable and export compatibility against the written brief. A human verifies product/advertising claims.
• Record total credits/currency, wait, editor repair time and accepted shots. Cost per accepted shot is comparable only with equivalent tasks and rights.
• Export into the real editing tool and finish a draft before buying annually. If nothing passes, use conventional footage or revise the brief; do not invent a winner.

Failure cases and method#

Altered packaging, inconsistent identity, unapproved likenesses, unusable audio, inaccessible endpoints and missing export features can all defeat the workflow. A vendor demo or advertised maximum duration does not resolve them.

Historical context: Runway launch, annual documentary framework. See editorial standards. October 15 removes unsupported app-history and superiority claims, adds a scoped shortlist and separates proposed tests from results.