Skip to content

Sora 2 Prompt Mistakes: Archived Writing Guide

Sora 2 Prompt Mistakes: Archived Writing Guide

Availability notice

Sora is no longer available. The web and app ended on April 26, 2026, and the API ended on September 24, 2026, according to OpenAI’s discontinuation notice. This is a historical writing reference, not instructions for accessing an active Sora service.

These examples illustrate ways to describe a scene. They are editorial suggestions, not benchmark results, and can be adapted to another available video tool.

Too many events

A short clip is easier to evaluate with one clear subject and action. Instead of requesting a chase, conversation, transformation, and location change together, draft a separate shot for each event.

Example: A courier stops beneath a streetlight, looks at a paper map, and turns toward the alley. The camera remains in a medium shot.

Incompatible camera directions

A locked shot and an orbiting camera cannot describe the same continuous movement. Pick one primary instruction, then add a second shot only if the tool supports that workflow.

Example: The camera slowly tracks beside a cyclist along a coastal road. The horizon remains level.

Prompt words mistaken for settings

Writing a resolution, frame rate, or duration in prose does not make the service support it. Use the video tool’s actual controls for technical settings. Keep the prompt focused on composition, movement, light, and mood.

Ambiguous dialogue

Identify who speaks and keep the line short. Separate speech from sound effects. Review the generated audio instead of assuming every line will be rendered exactly.

Example: A shopkeeper says, “Good morning.” A doorbell rings as the customer enters. There is no background music.

Depending on perfect lettering

If a project needs an exact brand name or subtitle, plan to add it during editing. Text inside a generated scene may need correction. Test any requirement for precise lettering before making the whole shot depend on it.

Skipping review

Check the subject, motion, continuity, sound, and unintended text before using a clip. Preserve the settings of an acceptable result and change one instruction at a time when revising it.

Read the archived model guide or browse reusable examples.