Sora 2 for ads: what changed and what to use instead

What Sora 2 was good at for ad work, and its status now

Sora 2 was OpenAI's video model for "richly detailed, dynamic clips with audio from natural language or images", per the archived guide (OpenAI video generation guide). It is no longer available, so everything in this section is historical.

OpenAI's help center states: "The Sora web and app experiences were discontinued on April 26, 2026. The Sora API will be discontinued on September 24, 2026" (OpenAI Help Center). The API guide now reads: "No one-to-one replacement API is available." The deprecations page lists sora-2, sora-2-pro and their snapshots with a shutdown date of 2026-09-24 and no recommended replacement (OpenAI deprecations).

For ad work, two specs stood out: 20-second single generations and reusable non-human characters. The face restrictions limited it for human-presenter UGC.

One listing to watch. Higgsfield's pricing page still shows "Sora 2", "Sora 2 Pro" and "Sora 2 Max" rows (Higgsfield pricing), although OpenAI shut the API down on 24 Sep 2026. Do not plan a production on those rows.

Which ad formats Sora 2 suited

Sora 2 suited non-human and stylised formats, because its published character feature covered reusable non-human subjects while human faces in input images were rejected. The table is guidance from the archived capabilities, not test results, and each row names where to make that format now.

Human-presenter UGC, podcast and talking-head formats did not fit Sora 2's published restrictions on human likeness. Those formats move to tools with published lip sync, such as Kling or Higgsfield's lip-sync models.

Workflow: moving a Sora 2 ad workflow to another tool

Export what you made, then rebuild the workflow on a current tool with matching specs: script, shot list, references, generation, edit. The script and shot list carry over unchanged; only the generation step moves.

1. Export your Sora content. OpenAI's help center points to a content export at sora.chatgpt.com/sunset (OpenAI Help Center). Save finished ads, raw clips and any character reference images.
2. Pick a replacement by spec. Match the Sora feature you relied on to a published equivalent (table below).
3. Re-shoot the shot list. Keep your shot durations. If a shot was 20 seconds on Sora, it fits one Seedance 2.5 or Kling 4.0 generation; on Kling 3.0 or Seedance 2.0, split it into two shots of 10 seconds or less.
4. Rebuild characters. Recreate each non-human character as a reference sheet (front, side, three-quarter) and save it as an element or reference set in the new tool.
5. Regenerate and compare. Generate the hook shot first and compare it against your Sora version before regenerating the rest.
6. Edit. Join clips, add captions, music, the end card, and any price or offer as editor text.

Prompt structure that carries over from Sora 2

A Sora-style prompt that describes one shot in plain language carries over to Seedance, Kling and Veo with small changes: shot type, subject, setting, action, camera, sound. Shorten each prompt to the new tool's clip length.

The templates below are written for this guide and work on current tools. Square-bracket items are placeholders.

UGC selfie hook (for a tool with lip sync)

Vertical selfie shot, phone at arm's length, [presenter reference].
Setting: [location], [time of day] light.
The presenter looks into the lens and says: "[hook line]"
Small natural movements, handheld shake, no music.
Duration: [clip length supported by the tool].

Product hero shot

Close-up of [product] on [surface], [background colour].
Slow push-in as [one physical action, e.g. lid opens / steam rises].
Ends on the label, centred and readable.
Sound: [single product sound]. No dialogue.
Leave empty space at the top of frame for text.

Podcast two-shot (one host per shot, cut together)

Medium close-up, [host reference] at a studio microphone with headphones.
[studio set details] behind, warm practical lights.
Host says: "[one line of the conversation]"
Static camera. Audio: close mic voice, room tone.

Animated explainer scene (non-human character)

[animation style, e.g. clay / 3D cartoon] scene.
[character reference sheet] stands next to [problem object] and shakes its head.
It swaps the object for [product] and turns to camera.
Character says: "[one-sentence benefit]"
Camera: slow push-in. Lighting: [soft / bright], palette [colours].

Brand-film establishing shot

Wide cinematic shot of [location] at [time of day], [weather].
[one moving element] crosses the frame.
Slow aerial move forward, revealing [landmark or product in scene].
Audio: [ambient bed]. No dialogue. [grade, e.g. warm film look].

Limits, terms and policy notes

The main limit is that Sora 2 cannot be used for new work. Terms still matter for ads you already made with it.

Ownership of existing output. OpenAI's Services Agreement says the customer "owns all Output" and "OpenAI hereby assigns to Customer all OpenAI's right, title, and interest, if any, in and to Output" (OpenAI Services Agreement). Ads exported before the shutdown remain yours to use under those terms.

Pricing. No longer published; Sora does not appear on the API pricing page (OpenAI API pricing). The help center says "Any purchased ChatGPT/Sora credits can still be used for Codex."

Likeness. Sora 2 blocked real people and human-likeness character uploads by default. The replacement tools allow human presenters, so the responsibility moves to you: TikTok's ad policy does not allow misuse of a public figure's likeness, and lists "Making the primary subject say something they didn't actually say, using AI voice-cloning" as significant AI modification (TikTok ads policy).

Platform labels. Any ad you rerun from a Sora export, or rebuild on another tool, needs AI labelling. Meta labels ad images "created or edited using third-party AI tools with AI info" (Meta), and TikTok rejects or restricts undisclosed AI-generated ads. See AI-generated content labels on Meta and TikTok ads.

Sora 2 ads and the one-script, every-format method

The shutdown is a reason to keep scripts separate from tools. If a winning script is stored as beats, lines and a shot list, it can be regenerated on any model in any format when a tool disappears.

Store the script, voice track and character reference sheets outside the tool, and treat each model as replaceable. The full method is in one winning script, made in every ad format.

Anglr pays AI creators, editors and agencies on performance to remake winning ad scripts in every format. Anglr provides the scripts and angles, and every video gets a QC and compliance review before it goes live. If you make full ads with current tools and know they perform, see how it works for creators.