Script to Video AI in 2026: Models After Sora

For script-to-video work, break the script into shots, select an available model for each shot’s duration and references, then assemble and review the approved clips. Sora 2 is retired: its September 24, 2026 API shutdown date has passed. A new pipeline needs a current route rather than a plan to migrate before September.

This guide keeps six practical uses—hooks, B-roll, narrative shorts, transitions, reference chaining, and multilingual versions—and separates model limits from the editor workflow. It does not establish a head-to-head quality winner or a universal per-second price.

Last verified: 2026-10-28. Sources below establish documented capabilities and inspected integration limits, not newly completed generations or a deployed-production test.

Model choice after the Sora shutdown

Sora’s retirement changes current pipeline choices, not the value of assets you already exported. Keep those assets and rebuild new-generation dependencies on an available route. Transfer the shot brief and references, then inspect the new output rather than expecting the same visual result.

For Seedance, name the version: full 2.0, Fast, Mini, and 2.5 differ in duration and resolution. Full 2.0’s 15-second ceiling is not a ceiling for every Seedance model. Seedance 2.5 documents up to 30 seconds of base generation; that is not evidence that it will match the quality or motion of an old Sora shot.

For Veo, name the provider route. The retired Gemini Developer API preview and the Cloud GA model are different endpoints. ChatCut’s inspected tool exposes Gemini Omni, not Veo; a vendor model card does not establish a working ChatCut adapter. Likewise, ChatCut’s Kling new-generation route is currently suspended, which is separate from Kling’s native capabilities.

Budget the request you can actually submit

The older blanket claims that Seedance costs one-tenth as much as Sora, or that a 30-second Fast clip costs a fixed dollar amount, did not establish matched settings or valid single-request durations. They should not be used as current quotes.

In ChatCut, read the estimate for the exact model, resolution, aspect ratio, duration, and reference-video input. Reference-video duration can affect billing. Credits policy is the current rate owner; generation credits explains estimates and ledger checks.

For an external API, confirm its billing unit and settings. Google Cloud’s fetched Veo table lists video-only and audio variants, but the displayed unit needs confirmation against the applicable SKU before converting it into a per-second estimate.

A useful planning quantity is total generated seconds, including retries. For example, twenty five-second B-roll requests total 100 generated seconds before any retries. Apply a current quote to those settings; do not assume those twenty requests cost the same as one 100-second request, which the routes discussed here do not support.

Six use cases, six workflows

Hook clips. Write a three-to-five-second visual brief. Compare candidates on the actual motion, framing, text, and sound you need. Sora is no longer an option for new generation; choose an available route and inspect the result before committing to a series.

B-roll insertion. Mark the script’s visual cues, use short requests within the selected model’s limit, and keep the reference/style instructions consistent. Seedance 2.0 Fast or Mini may be worth comparing when 480p/720p fits the task. Check their current quote rather than treating “Fast” as a guaranteed time-to-result or fixed discount.

Narrative shorts. A 15-to-60-second finished scene can contain several shorter generated clips. Seedance 2.0 can also follow multiple ordered shots inside one 4–15-second request. Neither approach guarantees character consistency. Reuse approved references and inspect identity, geography, and continuity across every cut. If you need a single request longer than 15 seconds, evaluate Seedance 2.5’s separate limits.

Animated transitions. Specify the start, end, and intended transformation. Use first-and-last-frame mode where supported, or generate a transition as a separate asset. Review the intermediate frames; supplying endpoints does not guarantee every frame between them will be usable.

Reference-image chaining. Generate or import a still, approve its composition, pass it to an image-to-video route, then place the approved clip in the editing project. This makes visual decisions before video generation. It is a workflow method, not proof that image-to-video always beats text-to-video.

Multilingual versions. Approve the main cut first, then make the voice, lip-sync, and caption versions you need. Budget those steps separately from video generation. Inspect proper names, timing, and the final mix in each language.

Where script-to-video belongs in a production pipeline

Use generated clips where they serve a concrete script cue: a hook, an illustrative cutaway, an atmospheric establishing shot, or a transition. Combine them with filmed material, screen recordings, titles, and sound according to the project’s needs.

A full narrative made from generated footage needs shot planning and continuity review. This guide has no measured platform-ranking result or controlled evidence that generated content alone causes a particular enforcement outcome. Evaluate the finished story and the platform’s current publishing requirements directly.

The ChatCut workflow

Inside a ChatCut project, use AI image generation or import an approved reference, then choose a currently available AI video generation route. Check the input role, duration, resolution, and estimate before generating. Review the returned clip and place it on the timeline alongside the rest of your material.

Full Seedance 2.0 generation currently reaches up to 1080p in the inspected ChatCut route. That does not impose a universal 1080p final-export ceiling: Desktop local export and Web export have their own renderer rules. See exporting and Web export limits. Upscaling and selecting a larger export frame do not demonstrate that the original model synthesized extra detail.

Check plans and credits for access and the current estimate for cost. This article does not promise that a free editing plan includes a particular generation or reuse an old flat credit figure as a current tariff.

For a longer project, the AI filmmaking workflow covers assembly, while text-based editing explains the transcript layer.

Five questions worth a direct answer

Sora 2 or Seedance for hero hooks in 2026? Sora 2 is retired. Evaluate an available Seedance route or another current model using the actual shot brief. This article does not establish a quality winner.

What does a 30-second AI hero shot cost? Get a quote for the route and settings first. Seedance 2.0/Fast/Mini cannot produce a 30-second base clip in one request; assemble shorter requests or evaluate Seedance 2.5’s 4–30-second range. Retries, reference video, and later finishing add to the budget.

Can AI replace my whole video production? It can generate assets, but the final production still needs decisions about story, timing, continuity, sound, and output review. The appropriate mix depends on the project.

What’s the September 24, 2026 Sora deprecation? OpenAI lists that date as the shutdown of Sora 2, Sora 2 Pro, their listed snapshots, and the Videos API. It is a past retirement date as of this update, not a future migration window.

Worst script-to-video mistake in 2026? Budgeting or planning an entire cut before checking that the chosen route supports its duration, inputs, and output settings. Start with a shot list and an approved sample, then scale the workflow.

Try the image-to-video workflow in ChatCut, with the selected model’s limits and estimate visible before submission.