agentleFS
inreels-aiinreels-ai/inreels-skills/skills/ugc-ad/SKILL.md

ugc-ad

Skill1 starsChanged 11 days agoMITView on GitHub

Make a UGC-style video ad from a product image — creator, spoken script, native audio, rendered MP4 — using the inReels MCP server. Use when asked for a UGC ad, creator ad, testimonial ad, talking-head product video, or a video ad for Meta/TikTok/Instagram.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 2 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the image is cluttered, lit badly, or the packaging is unreadable, say so now.

    line 32
  2. Very specific:

    If you cannot make the claim in the budget, do not submit a long script.

    line 63
  3. Distinctive:

    A product image the viewer cannot read: If the label is illegible at thumbnail size it is illegible in the feed.

    line 120

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

UGC ad

A person holding your product, talking to camera, in a video that looks like it was filmed on a phone. The thing that actually sells on Meta and TikTok.

Requirements

inReels MCP connected. Verify with get_account.

claude mcp add --transport http --scope user inreels https://www.inreels.ai/api/mcp

create_ad_video needs a paid plan. If the user is on free, say so before writing a script rather than after.

You need a product image first

The ad is built around it — this is not optional and there is no text-only path.

  • The user gives you an image URL → use it.
  • The user gives you a file or a pasted image → upload_image with base64, and use the URL it returns.
  • The user gives you only a product page → ask for the image. Do not generate a fake product; the whole point is that their actual product appears.

If the image is cluttered, lit badly, or the packaging is unreadable, say so now. Every frame of the ad inherits it.

Method

1. Angle before script

One ad makes one claim. Establish, from the product and whatever the user told you:

  • Who it is for, specifically. "Busy people" is not a who.
  • The friction they feel today, in their words, not the brand's.
  • The angle — the single claim this ad makes.

A landing page's own headline is usually the weakest available angle, because it was written to satisfy everyone. Its FAQ and objection-handling tell you more about the real hesitation. Aim there.

2. Write to the word budget, or don't write at all

The script sets the length, and the budget is tight:

LengthWords
5sup to 10
10sup to 20
15sup to 30

Over budget and the line gets clipped mid-sentence — the model does not speak faster to fit. Cut words rather than hoping.

If you cannot make the claim in the budget, do not submit a long script. Either go up a length, or omit script entirely and pass a brief instead, which lets the model write its own line to fit.

A UGC line is spoken, not written. Contractions, one idea, no brand voice. "I stopped buying the expensive one" beats "Discover our premium formulation".

3. Pick the creator

list_actors returns the roster. Pass creator_actor_id for one of ours, or creator_image_url for the user's own photo.

Omit both and you get a product-led ad — no person, just the product in motion. That is the right choice for a product that demonstrates itself, and for categories where a presenter reads as an advert.

4. Quote, then render

quote_ad_video { "product_image_url": "…", "script": "…", "variations": 3 }
create_ad_video {
  "product_image_url": "…",
  "creator_actor_id": "<from list_actors>",
  "script": "<within budget>",
  "look": "natural",
  "aspect_ratio": "9:16",
  "variations": 3
}

look: natural reads as real UGC and is the default worth keeping; raw_iphone is deliberately rough; commercial is polished and stops looking like UGC, which is usually the opposite of what was asked for.

Ask for variations. Ads are tested in sets, not shipped one at a time. Three angles beat one angle rendered beautifully. Each variation is charged separately, so quote the set and say the total before submitting.

5. Poll and hand over

get_run_status { "run_id": "…", "wait": true }

Minutes, not seconds. Keep polling until completed, then give the user the MP4 URLs. Never hand over a URL from a run that has not reported completed.

Signed URLs expire — refresh_asset_url mints a new one.

What makes these fail

  • Script over budget. The single most common failure. Count the words.
  • A hook that is a brand statement. The first two seconds decide the ad. Open on the friction or the result, never on the company.
  • A product image the viewer cannot read. If the label is illegible at thumbnail size it is illegible in the feed.
  • One variation. You cannot tell what works from one ad.
  • Asking for a face when the product is the story. Omit the creator.

Spending rules

  • create_ad_video spends credits and needs a paid plan.
  • Quote before you spend; the quote is exactly what the call charges.
  • Failed runs refund in full.
  • Credits are charged on completion by the pipeline, not at submit.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.