Social Media Metadata Generator

Give it a video topic or a full script and it returns a complete publishing metadata pack: several title options, an optimized description/caption, hashtags, SEO keyword tags, and a pinned-comment idea. Everything is tuned to one platform at a time (YouTube long-form, YouTube Shorts, TikTok, Instagram Reels, or X/Twitter) so the character limits and hashtag counts match where you're actually posting. It's meant for people running a content pipeline who want to stop hand-writing titles and tags for every upload.

How it works

Each item goes through a single chat-completion call with a platform-specific prompt that enforces that platform's title length, description limit, and hashtag count. The response comes back as structured JSON, then gets cleaned up (hashtags normalized to #nospaces, tags stripped of #).

Input

You need either text or sources. Everything else has a sensible default. An OpenAI key is required for real generation because it runs on an AI model. If you run it with no key, the actor returns one clearly-labeled sample record (_sample: true) so you can preview the output shape — add your key for real, AI-written metadata.

Field	Required	Notes
text	one of text/sources	The topic, title idea, or full script to generate metadata for.
sources	one of text/sources	Array for batch runs. Strings, or objects with title/script/scriptText/text/selftext. Pipe the Script Rewriter or a Reddit scraper straight in.
platform	no	youtube, youtube shorts, tiktok, instagram reels, or twitter. Default youtube. Drives title/desc limits and hashtag count.
titleCount	no	How many title variations to return, 1 to 15. Default 5.
keywords	no	Comma-separated SEO keywords to weave in where natural.
language	no	Output language. Default English.
openaiApiKey	yes	Your OpenAI key (bring-your-own). Marked secret. Can also be set via OPENAI_API_KEY.
model	no	Chat model, default gpt-4o-mini.
baseUrl	no	OpenAI-compatible base URL if you're routing through something else.

Output

One dataset record per input item. Each record has platform, titles (array), description, hashtags (array, #-prefixed, no spaces), tags (array of bare SEO keywords), and pinnedComment. Records also carry ok and the truncated source; if a single item fails the run keeps going and that record gets ok: false with an error.

Example

{
  "text": "A 60-second video explaining why cats knock things off tables, with funny examples.",
  "platform": "youtube shorts",
  "titleCount": 5,
  "keywords": "cat behavior, funny cats, pet tips",
  "language": "English",
  "openaiApiKey": "sk-..."
}

Pricing

$0.015 per metadata pack (pay per result, no subscription). You also pay your own OpenAI usage on your key.

Notes

A small titleCount (5) is a fine default. Pushing it to 15 spends more tokens for variations that start to repeat themselves. Pick the right platform before running, since switching it changes the limits the model writes to, not just a label.