Sora Prompt Guide: How to Write Video Prompts—and the API Status

Important API update: OpenAI discontinued the Sora API on September 24, 2026. Its Sora web and app experiences ended earlier, on April 26, 2026. That means you can use the prompt-writing guidance below as a creative framework, but you cannot follow it with a currently available OpenAI Sora API request. Don’t build a new integration around archived Sora API examples. (OpenAI Help Center)

The short answer: describe what should be visible

A useful video prompt makes the intended scene easy to picture. Start with the subject and action, then add the setting and any camera, lighting, mood, or style details that matter. Be specific about requirements; leave less important choices open if you welcome creative interpretation.

This is a practical writing method, not an official Sora syntax or a promise that a model will follow every instruction. OpenAI’s Sora 2 prompting guide makes the distinction between detailed shot direction and creative freedom; it describes a prompt as creative direction, not a contract. The guide is now archived and may describe outdated models or APIs. (OpenAI’s archived Sora 2 prompting guide)

A simple prompt-building framework

Use these fields as a worksheet—not as required syntax:

1. Subject: Who or what is the shot about?
2. Action: What visible movement or event happens?
3. Setting: Where and when does it happen?
4. Camera: What framing or movement would help show it?
5. Look and light: What visual style, color, or lighting matters?
6. Sound or dialogue: Include it only when relevant and supported by the video tool you’re using.
7. Priority: Which one or two details must not be lost?

You don’t need to fill every field. Choose details that clarify the scene rather than adding adjectives or instructions that don’t change what the viewer would see.

Reusable draft:

[Subject] [visible action] in [setting]. The camera [framing or movement]. [Lighting and visual style]. [Optional sound or dialogue]. Prioritize [essential detail].

Example: turn a vague action into a filmable moment

Vague: “Someone lights a candle.”

More specific: “Close-up of hands striking a match and lighting a candle; wax melts slowly as the flame steadies in soft focus.”

The revised version says what the viewer sees: hands, a match, a flame, melting wax, and a close-up. This illustrative contrast comes from Ramdam’s third-party tutorial; it is not a controlled test showing that one wording reliably performs better. (Ramdam’s Sora 2 prompt guide)

Example: give a scene a clear visual identity

Vague: “A cozy room.”

More specific: “A small room lit by a warm lamp, a blanket on a couch, and a cat sleeping.”

Ramdam’s tutorial uses this example to replace a general mood word with visible scene elements. The choice of whether to add more details depends on the brief: if the exact room layout is unimportant, leaving it open gives the model room to interpret. (Ramdam’s Sora 2 prompt guide)

How much detail should you include?

Use more detail when a choice is important to the brief and less when variation is welcome. A short prompt can be a good starting point for exploration. A more directed prompt can communicate the shot you have in mind—but piling on low-priority details can make it harder to see what matters most.

For example, Ramdam’s tutorial gives this short prompt, which combines a visual style, subject, location, and spoken line:

In a 90s handheld interview, an old Japanese fisherman stands by the dock and says, “The sea was wilder back then.”

The prompt leaves other choices unspecified, including the weather, camera angle, outfit, and emotional tone. That can be appropriate if you’re open to interpretation. If a particular weather condition or angle is essential, say so. The example and its explanation are from a third-party tutorial; they do not guarantee what a model will infer. (Ramdam’s Sora 2 prompt guide)

A useful decision rule is:

• Specify details that are requirements: the main action, important objects, or the visual treatment the project depends on.
• Leave open details that are flexible: incidental background choices or creative variations that don’t affect the brief.
• Simplify when a prompt contains many competing instructions. Restate the central action and the most important visual priorities.

OpenAI’s archived Sora 2 guide also recommends balancing specific shot direction with creative freedom. Because the guide is archived, use it as historical prompt-writing guidance—not as documentation of a currently available Sora service. (OpenAI’s archived Sora 2 prompting guide)

A practical refinement loop

When a generated clip misses the brief, change the instruction that addresses the mismatch rather than rewriting everything at once:

1. Compare the output with the intended shot. Name the specific difference: wrong action, unclear subject, unwanted framing, or missing scene detail.
2. Choose one priority to adjust. For example, make the action more explicit or state the camera framing.
3. Revise that part of the prompt. Keep details that already express the brief; remove distractions if the prompt has become crowded.
4. Generate again and compare. Treat the result as another variation, not proof that the wording guarantees a specific outcome.

For a practical prompt notebook, record the version of the prompt, the change you made, and what changed in the result. This makes comparisons easier. It is a workflow suggestion, not a claim that revisions will reliably improve every generation. Ramdam’s tutorial similarly recommends small adjustments and observing which wording helps or confuses; it does not establish a fixed number of attempts or guaranteed success. (Ramdam’s Sora 2 prompt guide)

Can you use this guide with the Sora production API?

No—not with OpenAI’s discontinued Sora API. OpenAI’s Help Center says the API was discontinued on September 24, 2026, and that the Sora web and app experiences were discontinued on April 26, 2026. As a result, this guide cannot provide a current Sora model name, endpoint, request code, or prompt-to-video API workflow. (OpenAI Help Center)

OpenAI’s developer site still lists a Sora 2 prompting guide, but the guide itself is archived and warns that it may reference outdated models or APIs. Its historical parameter examples should not be treated as instructions for a live production integration. (OpenAI’s archived Sora 2 prompting guide)

If you are choosing another video-generation service, check that provider’s current documentation for its supported model, input format, generation and retrieval steps, output controls, and availability before adapting this prompt framework. Those details are provider-specific; this article does not verify another service’s API or capabilities.

Bottom line

Build prompts from visible scene direction: subject, action, setting, and only the camera, look, or sound details that matter. Be precise about requirements, leave nonessential choices flexible, and revise one mismatch at a time. But do not treat archived Sora API material as usable production instructions: OpenAI discontinued the Sora API on September 24, 2026. (OpenAI Help Center)

Sources

What to know about the Sora discontinuation — OpenAI Help Center

Sora 2: Prompting Guide — OpenAI Developers (archived)

Sora 2 prompt guide: How to write better prompts for AI video generation — Ramdam