How many of 10 distinct objects will Sora depict, given a 100% visual prompt that doesn't include any of the objects?
Basic
8
Ṁ92
2026
43%
chance

I want to test Manifold's skills at "visual prompt engineering!" Once Sora is released publicly, people can start competing to come up with visual prompts that will cause Sora to generate as many as possible of these 10 objects:

  1. Rubber duck

  2. Comet

  3. Abacus

  4. Bicycle helmet

  5. Bridge

  6. Lemon

  7. Chessboard

  8. Tuba

  9. Wristwatch

  10. Rose

At the end of February 2026, this market will resolve to X%, where X is 10 times the greatest number of these objects that a visual prompt caused Sora to depict in a single video.

By "visual prompt," I mean some combination of images or video. The prompt must not include sound, text, or any other medium. The prompt must not contain text in any form, and it must not depict any of the objects listed above, in whole or in part.

The prompt must be part of the final video, and it must contain no more than two sections of contiguous frames. For instance, you can make Sora start the video with one clip and end the video with another clip, but you can't interleave 10 clips into the video, alternating back and forth between clips from your prompt and clips generated by Sora.

I will personally test Sora with any prompt submissions that I think are reasonably likely to work, using prompting instructions from the submitter. I will try each prompt up to 5 times. The generated video must be one minute or less. I won't spend more than $40 total to run this contest; if it looks like this could be a problem, I'll figure out how to give as many people as possible a fair shot given my budget.

I'll post the results to Manifold, so people can check my work. If there's any ambiguity about whether an object appeared in a generated video, I will put out a poll; at least 80% of respondents must answer that the object appeared for it to count.

Future versions of Sora or other text-to-video models released by OpenAI may also be used. If I can't run any prompts by the close date, this market will resolve N/A. I will not trade in this market.

Get
Ṁ1,000
and
S3.00
Sort by:

Was it confirmed that sora does image to video? I think it might be text-to-video only.

@ProjectVictory It can animate images and extend videos. See their research report: https://openai.com/research/video-generation-models-as-world-simulators

Related questions

© Manifold Markets, Inc.Terms + Mana-only TermsPrivacyRules