Which AI for which job

An AI model directory should help you make a production decision. A long list of names is useful only when it explains what the names refer to, which controls are actually available and when the information was checked. Otherwise, it can make incompatible tools look interchangeable and old instructions look current.

The 4AI media model directory is a curated reference. Its entries should be read as documented capabilities and editorial evaluation guidance. They are not a claim that every model has been tested by 4AI, a complete census of the market or a permanent ranking of quality.

Distinguish the model from the platform

A model is the generation or editing engine. A platform is the application or service through which you access it. One platform can offer several vendors' models, and one model can appear through several delivery routes. The controls, pricing, permissions and available versions may differ between those routes.

When comparing options, record both names. An instruction that works in a vendor's application may not describe a distributor's API. A feature listed for a model family may not be exposed in the plan or interface you intend to use. Confirm the route before treating the feature as part of a production commitment.

Read capabilities at the task level

Text-to-image, image editing, image-to-video, video editing and reframing answer different needs. A generative tool may accept a reference for appearance without offering a precise local edit. A video tool may create a new sequence without being able to preserve all elements of an existing clip.

Start with the work required. Do you need a fresh concept, a product-accurate adaptation, a controlled change or a sequence with consistent subjects? The profile should tell you which documented controls are relevant and what still needs to be tested. Do not assume that the longest feature list produces the most appropriate workflow.

Check the date and lifecycle

Model information changes. A family name can remain familiar while a version, endpoint or application feature is retired. A historical documentation page may still be accessible after the production route has closed.

There are current examples worth taking seriously. OpenAI's deprecation register records the retirement of its Sora 2 API route. Google's Veo guide distinguishes retired Gemini preview endpoints from separately documented access routes. Midjourney's version documentation explains a reference workflow change between generations. These are reasons to retain verification dates, not reasons to avoid AI production altogether.

Check lifecycle information again before an integration, a purchase or a delivery commitment. Keep approved creative assets and production decisions outside a single model so a change of engine does not force a complete restart.

Separate documentation from evidence of quality

A vendor can document a supported input or output. That is different from proving that the output will meet your brief. A promotional comparison is also different from an independent evaluation with a consistent method.

For a meaningful test, use the same references, task and acceptance criteria. Record the platform, version, date and attempts. Include rejected results and finishing work in the comparison. If the test covers only one product image or one shot, describe that scope instead of presenting the result as a general quality ranking.

Treat missing information honestly

An unverified entry should be labelled as research in progress. If a source cannot be inspected, a profile should not fill the gap with a confident specification copied from an unrelated platform. Where a capability depends on account access or a preview programme, state that condition next to the claim.

The directory can still record the family in an unpublished backlog. Publication should follow verification and useful editorial work. A page with little more than a logo, a version number and a sales button is not a substitute for a practical guide.

Does this directory list every available engine?

No. It covers a selected set with explicit verification states. New families can be added when their primary documentation can be checked and the entry helps a real production decision.

Does a model profile mean 4AI uses that engine?

No. A research profile is not a record of production use. A project proposal should identify the actual method and relevant deliverables for your brief.

How should a business use the directory?

Make a shortlist, then test a representative asset or shot. Send 4AI your brief for a scoped recommendation, or explore AI video and visual design.

Sources and verification

Checked on 27 October 2026: OpenAI deprecations, Google Veo lifecycle notice, Midjourney versions and Runway's multi-vendor catalogue.