Social Metadata Generator - Titles, Hashtags & SEO Tags

Give it a topic, a title idea or the whole script and get back the titles, caption, hashtags and tags for your video: several title options, a description, hashtags, 10 to 15 keyword tags and a pinned comment, written for the platform you picked.

Two things to know up front. You bring your own OpenAI key, so the writing is on your account and your bill. And the platform rules are asked of the model rather than trimmed afterwards, so a title can still come back longer than YouTube will show. Read them before you post.

Input	A topic, a title idea or a full script, one at a time or as a batch
Output	One pack per source: title options, description, hashtags, 10 to 15 tags and a pinned comment
Ceiling	One platform per run, source text read up to 6,000 characters
Account needed	Your own OpenAI key
Price	$15.00 per 1,000 packs ($0.015 each) on the Free plan, a little less on paid plans. The free plan's $5 a month covers about 330

🔍 What Social Metadata Generator does

Every platform wants something different from the same video. YouTube rewards a title that reads like a search, TikTok wants the first three words to land, Reels leans on the caption, X has almost no room at all. Pick the platform in platform and the pack is written for that one.

A pack is five things: titles with as many options as you asked for, a description you can paste as the caption, hashtags already normalised with the # and no spaces, tags as bare keywords for the YouTube tag box, and a pinnedComment to drop under your own video.

Run it with sources to do a batch. One pack per item, all on the platform you chose.

📋 What data you get from each metadata pack

What you get	Field
Title options, as many as you asked for	titles
A caption or video description	description
Hashtags with the # already on	hashtags
Bare keyword tags for the YouTube tag box	tags
A comment to pin under your own video	pinnedComment
The platform it was written for	platform
The start of what you sent, so you can tell packs apart	source

▶️ How to write titles and hashtags from a topic or a script

1. Open Social Metadata Generator and click Try for free.
2. Paste your topic or script into Topic / script.
3. Pick the Platform. This is the one setting that changes the output most.
4. Put your key into OpenAI API key (BYO). Without it you get the sample pack instead.
5. Click Start, then take the pack from the dataset or read it from the API.

💰 How much does it cost to write titles and hashtags for a video?

$15.00 per 1,000 packs, which is $0.015 per pack on the Free plan. Paid plans pay a little less: $14.80 on Starter, $14.60 on Scale and $14.40 on Business. On the free plan, the $5 Apify gives you each month covers about 330 packs.

One source in, one pack out, one charge. The sample pack and any item the model could not write go to the key-value store instead of the dataset, so neither is billed here.

Your OpenAI usage is separate and goes to OpenAI on your own key. gpt-4o-mini is the default because it is cheap for this kind of short writing job.

📥 What you give it

{
  "text": "A 60-second video explaining why cats knock things off tables, with funny examples.",
  "platform": "youtube shorts",
  "titleCount": 5,
  "keywords": "cat behaviour, why cats do that",
  "language": "English"
}

Field	Default	What it is
text	none	One topic, title idea or script. The form opens with a sample topic in it.
sources	none	A batch. Strings, or objects carrying title, script, scriptText, text or selftext.
platform	youtube	youtube, youtube shorts, tiktok, instagram reels or twitter. One per run.
titleCount	5	Title options to write. 1 to 15.
keywords	none	Comma separated terms to work in. Useful when you already know what you rank for.
language	English	Output language. Write the language name in plain words.
openaiApiKey	none	Your own key. Held as a secret field, never written into the results.
model	gpt-4o-mini	Any chat model your key can reach.
baseUrl	none, which means OpenAI	Point it at any OpenAI-compatible endpoint instead.

An object in sources is read title first. If it has a title, that is what gets used and the script alongside it is ignored. When you want the whole script considered, pass it as a plain string.

📤 What you get back

Real output from a run with no key on it, so this is the free sample pack the actor writes rather than model output, which is why the text says so. A real pack has the same fields without _sample and note:

{
  "ok": true,
  "_sample": true,
  "note": "SAMPLE output ... Add your OpenAI key (openaiApiKey or the OPENAI_API_KEY env var) to generate real, AI-written metadata.",
  "source": "A 60-second video explaining why cats knock things off tables, with funny examples.",
  "platform": "youtube",
  "titles": [
    "A 60-second video explaining why cats knock things off table ... You Won't Believe #3",
    "The Truth About A 60-second video explaining why cats knock things off table",
    "A 60-second video explaining why cats knock things off table: A Quick Guide"
  ],
  "description": "[SAMPLE] Hook line about A 60-second video explaining why cats knock things off table...",
  "hashtags": ["#sample", "#demo", "#metadata", "#contentcreator", "#socialmedia"],
  "tags": ["sample", "demo", "metadata generator", "titles", "hashtags", "seo tags"],
  "pinnedComment": "[SAMPLE] Drop a comment below ... this is example engagement bait."
}

The title list and hashtag list are cut short above so the block stays readable.

Field	How to read it
titles	Your options, in the order the model wrote them. Pick one, they are not ranked.
description	The caption or video description, as one block of text.
hashtags	Already carrying the # with the spaces taken out, so they paste straight in.
tags	Bare keywords, no #. This is the YouTube tag box, not the caption.
pinnedComment	A comment to pin under your own video. Hidden in the default table view, so open the JSON or export to see it.
source	The first 1,000 characters of what you sent, so you can tell packs apart in a batch.

🧾 Reading the output

A dataset row is a finished pack, and every dataset row is billed. Two other things the actor writes never reach the dataset at all, so they never bill:

What	Where it goes	Billed
A finished pack	The dataset, ok: true	yes
The sample pack, when no key was supplied	Key-value store, SAMPLE_AND_NOTICES, flagged _sample	no
An item the model could not write	Key-value store, same key, ok: false with the reason	no

So if you ran a batch of 20 and the dataset holds 17 rows, the other three are in Storage under SAMPLE_AND_NOTICES with the reason attached. A run where every item failed still finishes green with an empty dataset, which is why the row count is the thing to check rather than the run status.

Without a key, a batch gets one sample pack, for its first item. With a key on the run and nothing in text or sources, the run fails outright rather than writing an empty dataset.

💡 What people use it for

• Writing the title, caption and tags for a batch of shorts in one pass instead of one video at a time.
• Getting five title options to choose from, which is more useful than one confident guess.
• Rewriting the same video's metadata for a second platform by running it again with platform changed.
• Keeping a house set of keywords in keywords so every upload leans the same way.

From a topic to the strongest title, in two steps:

1. Run this actor on your topic with titleCount set to 10.
2. Put the titles into Hook & Virality Scorer as texts. It sends them back best first.

🚧 What it does not do

• It does not enforce the platform limits. Lengths, hashtag counts and the 10 to 15 tag range are asked for, not trimmed afterwards. Check a title before you paste it.
• No trend or search-volume data. The tags are written from your topic, not from what is currently being searched.
• One platform per run. Two platforms means two runs, and two packs to pay for.
• It reads an object's title, not its script. Pipe rows in and you may be generating from a headline. Pass the script as a string when the body is what matters.
• Long scripts are trimmed. Roughly the first 6,000 characters are what the model sees.
• No retry. If your key hits OpenAI's rate limit mid-batch, the remaining items can fail one after another. They are not billed, and the reason is in SAMPLE_AND_NOTICES.
• No key, no writing. A run without one produces the sample pack so you can see the shape, and nothing more.
• It does not post anything.
• It is not an official tool of any platform or company named on this page. Dami's Studio is independent and is not affiliated with or endorsed by any of them.

🧭 Which video tool do you need?

If you want	Use
Titles, captions, hashtags and tags for a video	This one
A 0 to 100 score for each title or hook before you post	Hook & Virality Scorer
A thumbnail with a close-up face and a hook headline	AI Thumbnail Generator
A post or article rewritten into a short-form script	Story to Script Rewriter

❓ Questions people ask

Why do I need my own OpenAI key?

Because the writing happens on your account, at your model choice, and nothing about your script passes through a key of ours.

Can I use a model that is not OpenAI's?

Yes, if it speaks the OpenAI chat format. Put the endpoint in baseUrl and the model name in model.

Where is my pinned comment?

On the row. The default table view does not show that column, so open the JSON or export the dataset.

Will the titles be under the YouTube limit?

Usually, but it is not guaranteed. The length is requested from the model and not cut afterwards.

What language can it write in?

Whatever you put in language. Write the name of the language, not a code.

Can I call it from code or connect it to an AI assistant?

Yes. The API tab has ready-made code for Python, JavaScript and the command line. For Claude, ChatGPT or another MCP client, connect https://mcp.apify.com/?tools=fetch-actor-details,dami_studio/social-metadata-generator. Either way the run happens on your Apify account at the same price, and the writing on your own OpenAI key.

🆘 If something breaks

Open the Issues tab on the actor page. Send the run ID and the topic you used. If rows are missing, look in Storage under SAMPLE_AND_NOTICES first, because the reason is usually written there.