Claption

caption lab for AMD GPU tracks
Overall 8.7 / 10 ยท Ready
Judge note: this Vercel URL is a public UI preview. For automated Track 2 scoring and full video processing with system FFmpeg, use the container image ghcr.io/2cloud-s/claption:latest. The Docker path is the intended evaluation runtime.

Video input

sample-clip.mp4

Load one of the hackathon clips to preview the four-style caption pass. Judges should use the Docker image for the official run because it includes the full FFmpeg runtime.

Grounded facts

A person demonstrates a short activity in a controlled indoor scene while the camera stays focused on the main action.

Actions: demonstrates, moves through a short sequence, finishes the action

Uncertainty: Sample mode uses placeholder facts until a real video is processed by the Python pipeline.

Run history
sample-clip8.7

Four judged captions

7 frames

Formal

9.1

A person completes a brief indoor demonstration while the camera remains focused on the main action.

Objective summary with no humor.

Grounded and direct.

Sarcastic

8.5

A focused indoor demonstration unfolds, bravely proving that one clear action can indeed fill a whole clip.

Dry irony without inventing extra events.

Sarcasm is mild and safe.

Humorous-tech

8.7

The clip runs a single-action demo with excellent uptime and no visible stack trace.

Uses software humor while staying tied to the observed action.

Clear technical joke.

Humorous-non-tech

8.6

Someone gives a quick indoor demo and wraps it up before the room has time to get bored.

Everyday humor without technical references.

Accessible joke.