Implement computer vision solutions 3.1 Design and implement image- and video-generation solutions

A Tokyo agency writes Sora 2 (preview) prompts in Japanese, such as 'a nice ad for our…

Study guide skill: Implement a solution that generates videos from text prompts and reference media

Question B7-002

A Tokyo agency writes Sora 2 (preview) prompts in Japanese, such as 'a nice ad for our new drink', and gets inconsistent clips. Which two changes follow the Sora 2 prompt guidance? (Choose two.)

Select 2 answers.

  • A Mention a well-known celebrity spokesperson by name to make the clip more realistic
  • B Write the prompts in English or another Latin-script language ✓ Correct answer
  • C Combine every scene of the ad into one long prompt so a single job produces the whole storyline
  • D Name a famous animated character so the style is easy for the model to recognize
  • E Describe the shot type, subject, action, setting, lighting, and any camera motion, and keep each prompt to a single purpose ✓ Correct answer

Answer: B, E

The Sora 2 prompt parameter guidance asks for shot type, subject, action, setting, lighting and camera motion to reduce ambiguity, and a single purpose per prompt for best adherence. The best-practice section recommends English or other Latin-script languages for the best results. Copyrighted characters are rejected and real people, including public figures, can't be generated, so those prompts fail.

Deep dive

How it works

Sora 2 (preview) gives prompt guidance in two places, and both apply here.

The prompt parameter. The API parameters table describes the prompt as a natural-language description of the shot. Include:

  1. shot type (for example, a close-up or wide shot)
  2. subject
  3. action
  4. setting
  5. lighting
  6. any desired camera motion

All of this reduces ambiguity. And keep it single-purpose for best adherence. "A nice ad for our new drink" gives none of these, which is why the clips vary.

Best practices for prompts. Write prompts in English or another Latin-script language for the best video generation performance. The agency can still brief in Japanese and translate the final prompt.

For a multi-scene ad, generate one shot per job with its own focused prompt, then edit the shots together.

The other options, and when you'd use them

  • Name a celebrity spokesperson: Real people, including public figures, can't be generated, so the job fails.
  • One long prompt for every scene: That works against keeping each prompt single-purpose for best adherence. Generate scene by scene.
  • Name a famous animated character: Copyrighted characters are rejected.

Exam tip

Sora 2 prompts: English or Latin script; shot type, subject, action, setting, lighting, camera motion; one purpose per prompt. No real people, no copyrighted characters.

References

Last reviewed:

Ready to test yourself? The quiz asks this and every other AI-103 question, with instant feedback. Practice this in a quiz →

All Computer vision questions · AI-103 study guide

Thanks for your feedback!