---
title: AI Video Captioner
emoji: 🎬
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: mit
---

# AI Video Captioner

Paste a direct video URL (under 2 minutes) and get an AI-generated caption in
4 styles: formal, sarcastic, tech-flavored humor, and everyday humor. Each
caption is drafted, graded by an LLM judge, and refined before being shown.

## Required secrets

Set these under **Settings → Variables and secrets** on the Space (not in any
file in the repo):

- `FIREWORKS_API_KEY` (required) — used for scene analysis and caption generation
- `OPENROUTER_API_KEY` (optional) — used for audio transcription; if omitted,
  captions are generated from visual frames only

## Optional variables

- `SCENE_MODEL`, `CAPTION_MODEL`, `JUDGE_MODEL`, `REFINE_MODEL` — model names
  on Fireworks (default to a `minimax-m3`-based setup)
- `MAX_RETRIES` (default 3), `MAX_REFINE_ITERATIONS` (default 1),
  `ENABLE_JUDGE` (default true), `MAX_DURATION_SECONDS` (default 120)

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference