proofof.video

What It Is

Proof of Video is a community gallery for AI/LLM-generated 2D/3D worlds. Its focus is not on generating videos or scenes itself, but on presenting side-by-side real video recordings of “interactive, explorable worlds” produced by different models and software pipelines under the same or similar prompts. The site explicitly states that it does not include linear videos or films, and focuses only on interactive 2D/3D worlds.

Core Capabilities and Use Cases

Its workflow is relatively transparent: a text prompt or scene description is first shared; contributors then use their own LLM models and 3D/generative software pipelines to produce results; they record the output, upload it to YouTube, and note the model and software versions; finally, they submit it via a GitHub Pull Request. Examples include Claude Opus 4.5, GLM 5, LocalGPT Gen v0.2.0/v0.3.0, and Rust/Bevy scene-generation descriptions. Typical use cases include comparing how models perform on tasks such as “heart outlines with spheres and cubes” or “use basic primitives to design a school,” providing reference material for AI 3D prototyping, world generation, and tool selection.

Pricing and Integrations

The website itself does not disclose pricing, subscriptions, or an API. Some example cards list experimental costs such as $0.5, $1, or $N/A, which appear more like per-generation or toolchain cost records and should not be treated as platform pricing. Integration is closer to an open-source community workflow: fork GitHub and submit a PR according to the guidelines. Videos are hosted on YouTube.

Pros and Cons

Its strengths are a clear comparison methodology and the requirement to document real videos, models, and versions, which helps reduce the opacity often found in marketing demos. The community submission mechanism also makes it easier to expand the dataset with more examples. Its limitations are that the current number of samples is limited, with relatively narrow coverage of models and scenarios. It is not an end-to-end generation tool, so users need to run their own models and 3D software. Information about privacy, the operating entity, and support services is also not disclosed.

Who It’s For and Access from China

It is suitable for AI 3D researchers, LLM application developers, game/virtual-world prototyping teams, and anyone who wants to observe differences in generation capabilities across models. For users in mainland China, the site relies on YouTube and a GitHub-based contribution workflow, so actually watching videos and participating in contributions may be partially restricted. Payment information is unclear. If you need to directly generate 3D assets or scenes, you may also want to look at alternative tools such as Meshy, Luma AI, Tripo, Spline AI, and Blockade Labs.