Implement computer vision solutions 3.1 Design and implement image- and video-generation solutions
A marketing team likes a Sora 2 clip but wants a small change (a different color for the…
Study guide skill: Implement workflows to edit generated videos
Question B1-033
A marketing team likes a Sora 2 clip but wants a small change (a different color for the car) without regenerating the whole scene from scratch. Which capability fits?
Select one answer.
Answer: B
Sora 2 can remix existing videos, making targeted adjustments instead of regenerating from scratch. A fresh generation can change everything, editing only the first frame doesn't carry through the clip, and Content Understanding analyzes media rather than generating it.
Deep dive
How it works
Sora 2 adds remix: the ability to change an existing video with targeted adjustments instead of regenerating it from scratch. You reference the ID of a successfully completed video and supply an updated prompt. The system keeps the original video's framework, scene transitions and visual layout while applying the change you asked for.
In the API, remix is a call like videos.remix with the previous video_id and a new prompt (the docs' example shifts the color palette). The new job reports remixed_from_video_id so you can trace which clip it came from. The create call also accepts an optional Remix_video_id, described as the ID of a previously completed video whose structure, motion and framing are reused.
The docs give one practical rule that fits this question: for the best results, limit each remix to one clearly stated adjustment. Narrow, precise edits stay closer to the source and are less likely to produce visual defects. "Make the car a different color" is exactly that kind of edit. If you need several changes, apply them as separate remixes.
Keep the job rules in mind too. Generation is asynchronous, you can run two video creation jobs at a time, and jobs are available for up to 24 hours. After that, you'd have to create a new job to generate the video again, so remix while the source job is still available.
The other options, and when you'd use them
- Run image inpainting on the first frame only: Editing one still frame doesn't carry the change through the rest of the clip. Remix works on the video as a whole.
- Use Content Understanding to rewrite the video: Content Understanding uses generative AI to process content such as documents, images, video and audio into a structured output you define. It analyzes media; it doesn't generate or edit video.
- Generate a new video with the same prompt and a different seed: A fresh generation starts from scratch, so the scene, motion and framing the team liked can all change. Remix exists to avoid that.
Exam tip
"Keep the clip, change one thing" means remix with the previous video ID and a narrow prompt. Do one change per remix for the best fidelity.
References
Last reviewed: