Related to: https://feedback.vrchat.com/feature-requests/p/tackle-the-misleading-world-thumbnails-issue I think the main problem with AI-generated world thumbnails is not AI itself, but the fact that a thumbnail can be completely disconnected from what the world actually looks like. Currently, the only reliable way to find out is to download the world and enter it. For larger worlds, that can mean spending significant time and bandwidth just to discover that the thumbnail was misleading. I propose adding automatically generated in-world screenshots to the World Details page, separate from the creator-provided thumbnail. When a world is submitted to Community Labs, or when a new public version is published, VRChat could load the world and automatically generate preview images from inside it. The important distinction is that these images would be generated by VRChat rather than uploaded by the world author, and clearly labeled as such. 〇 Capture locations At least one screenshot should always be automatically captured from the world's spawn point. World creators could then optionally specify several additional capture points in the SDK to showcase important parts of their world. ・Automatic screenshot from the spawn point: required ・Additional creator-defined capture points: optional This gives creators some control over presenting the interesting parts of their world, while still giving users an objective view of what they will actually see when they first join. These previews could be regular screenshots, or potentially 360-degree captures. 〇 Advantages This would not require users or moderators to decide whether a thumbnail is misleading. Unlike a rating or review system, it would also avoid issues such as subjective moderation or malicious downvoting. Creators could still freely use artwork, renders, logos, or AI-generated images for their main thumbnail. Users would simply have an additional way to compare the creator-provided promotional thumbnail with an automatically captured view of the actual world before downloading it. 〇 Possible manipulation Of course, this would not make deception completely impossible. A creator could build an area specifically for the automatic capture, or attempt to detect the capture process and manipulate the scene using shaders or Udon. However, the goal does not need to be perfect anti-cheat. Requiring at least one capture from the spawn point, combined with additional viewpoints or 360-degree captures, would make manipulation significantly more difficult than simply uploading an unrelated thumbnail. Even if the system can theoretically be bypassed, a clearly labeled "Automatically captured in-world" preview would still give users much more information about a world before they download it.
