The 2026 AI Video Review: Product Entry Points, Production Workflows and Availability Risk

Revised October 15, 2026: the July review is now a documentary decision framework. We removed the unverified seven-month Sora app narrative, claims of migration into ChatGPT, and unsupported model-winner or cost-leadership conclusions. It is not a hands-on benchmark. For a current, repeatable trial, use the video workflow guide.

Entry point comes before model name#

A model offered inside an app, through an API or as downloadable weights is a different purchase and deployment choice. Availability and permissions for one do not establish those of another. Google's January 13 announcement describes Veo 3.1 capabilities in the Gemini API; it is evidence for that developer surface, not a price or licence promise for every Gemini consumer plan.

A production workflow is more than a demo#

A small English-speaking product team needs approved source assets, repeatable shot instructions, reviewable exports and a route back to its editor. Runway's pricing page, checked October 15, lists model access, credits and storage by plan. Those are documented procurement inputs; they do not establish cinematic superiority, continuity or total usable-clip cost. Test those on your own permitted brief.

Treat Kling, Vidu, Hailuo and open-weight models as different follow-up investigations, rather than assigning a performance winner from product positioning. The historical comparison correction withdraws the unsupported speed, duration and price claims. A self-hosted workflow also needs an exact weight licence, hardware budget and operator; an open repository is not enough to infer unrestricted commercial use.

Availability risk: what the Sora evidence actually says#

OpenAI's official deprecation table lists removal of the Videos API and named sora-2 aliases/snapshots on September 24, 2026. No replacement is listed. That is sufficient to exclude that API/model scope from a current purchase. It is insufficient to date the standalone app's closure or claim a ChatGPT migration; those assertions are withdrawn. See the scoped directory record and corrected launch coverage.

A decision path#

• You need an app for approved product shots: inspect Runway's current plan and run the guide's export-and-correction trial before paying annually.
• You need programmatic generation: verify the exact live endpoint, quota, billing and acceptable-use terms. Do not select a retired Sora endpoint.
• You need local processing: first establish the exact weights, licence and hardware fit. This review has not verified a current local-model procurement package.
• You cannot verify regional access, client rights or source-asset handling: pause that candidate and ask the vendor for the missing terms before uploading client assets.

How we review#

This framework compares readable documentation and separates editorial acceptance criteria from vendor statements. No common-prompt video test, quality score or paid-plan procurement test was performed. Prices, version names and licences need their own current evidence; the framework does not rank models by unmeasured accuracy. See editorial standards.

Revision record: October 15 replaces unsupported app-history and superiority claims with a scoped availability correction and links to the current task guide. The original July 30 event date and URL are retained.