Claude Code content creation system

One command turns your context folder into ten finished LinkedIn posts, each with its own generated image, in about seventeen minutes.

10+

Finished posts per command

~17 min

End to end, mostly image generation

1

Command to start it

3:20

One line, create 10 posts, and about seventeen minutes later there is a dashboard with more than ten finished pieces of content in it. Not outlines. Finished: copy plus the visual, ready to publish.

The mix I use, which is just a sentence in the prompt:

• Four posts with a personal photo, chosen from your own library to fit what the post says
• Four AI infographics in your brand colours and type
• Two carousels, for Instagram or LinkedIn

That is roughly two to three weeks of posting, and every one of them is built on what is actually working in your niche this week rather than on what the model assumes.

This is the honest question and it has a specific answer. A chat window can write you ten posts right now. It cannot:

• See what is trending in your niche this week. That needs a scraper, and platforms block direct scraping, which is why Apify exists.
• Put your face in the image. That needs a photo library it can look through and an image model it can call.
• Hand you a dashboard. That needs somewhere to write files and the ability to build the page.

Claude Code does all three because it has a folder and can call other services. The difference is not writing quality. It is that one ends with text you then have to act on, and the other ends with the posts.

8:20

The project has folders you will mostly never open, and it is worth knowing what they are, because they are the reason the output is specific rather than generic. Think of it as the brain of a content person you just hired:

1. 1Open the template and start Claude Code

   Download the template, open the folder in VS Code (file, then open folder), open a terminal, and run claude. That is the whole install.

2. 2Add two API keys

   Open .env and paste two keys: Apify for scraping what is trending, and an image model key (i use kie.ai) for the visuals.

   APIFY_API_KEY=...
   KIE_AI_API_KEY=...

   Both take about a minute to get. These two keys are the difference between a chat that describes a post and a project that produces one.

3. 3Let it learn who you are

   Point it at your own profile and let it read your existing posts. On the account I set up in the video it pulled 600 posts and took about six minutes.

   This is what stops the output being generic. It is not learning to write, it is learning what you write about and how your audience responds.

4. 4Add your photos and one reference face

   Drop your photos into context/images/. Include one clear shot of your face named me.jpeg.

   Then ask it to catalogue them:

   i added images to context/images.
   rename them to something descriptive.
   use me.jpeg to identify which person is me.

   Fifty-four seconds later the filenames read orange-sofa.jpg, black-dress.jpg and so on. Now, when it writes a post, it can pick a photo that fits it rather than reaching for whatever is first.

5. 5Give it one style reference

   Put one image of the look you want into reference/. An infographic in your colours is enough: it takes the palette, the type and the general feel from that single picture.

   One good reference beats a page of description. The model is looking at the image.

6. 6Run the command

   create 10 posts. It scrapes what is trending in your niche, writes each post, generates the visuals, and opens a dashboard when it is finished.

   My default mix is four posts with a personal photo, four infographics and two carousels. All of that is just words in the prompt: change the numbers, ask for videos instead, whatever you need.

This is the detail people miss and it is worth its own section, because it is the difference between content that is obviously generated and content that looks like you made it.

Put one clear photo of your face in the images folder, named me.jpeg. Then ask Claude to rename everything else using it as the reference.

Two things happen. The model learns which person in a group photo is you, so it never picks an image where you are not the subject. And the filenames become descriptive, so when it writes a post about a move to a new city it can reach for the photo that fits instead of the first file in the folder.

You can see this working in the output. Every post comes with an image notes block underneath it, naming the exact file it chose and why:

## Image Notes
Personal photo: `oleg-selfie-at-desk-imac-night-city-lights-dubai-highrise.jpeg`
Vibe: casual, unpolished, matches the "imperfection" message.
Real desk, real setting, not a photoshoot.

And for a generated one it names the reference, the content and the style instead:

## Image Notes
AI infographic using `reference/infographic-ref-3.jpeg` (vertical flowing path).
Content: "From Voice to Published Post", vertical flow showing
Speak -> AI Processes -> Draft in Your Voice -> Quick Review -> Published.
Side-by-side time comparison: Old Way (90+ min) vs New Way (16 min).
Style: cream background, lime green accents, brand banner.

This is the part worth stealing even if you never use the project. A post and its image are one decision, so the system that writes the post should be the one that picks the picture, and it should have to say why.

The whole interface is one sentence, and every part of it is yours to change:

create 10 posts:
4 with my personal images,
4 as infographics in the reference style,
2 as carousels.

base them on what is actually trending in my niche this week,
and make each one useful to my audience rather than about me.

Want video instead of stills? Say so. Want twenty? Say so. Want them written for LinkedIn rather than Instagram? Say so. There is no settings screen to learn, which is the point.

• It is not fast per post. Seventeen minutes for ten. Fine when you walk away, annoying if you sit and watch.
• Skip the context step and it is generic. The whole value is in it having read your posts and your audience first.
• One weak reference image poisons every visual. The infographics inherit whatever you put in that folder.
• Ten drafts, not ten posts. Two or three will be genuinely good, most will need an edit, and one will be wrong about something. Read them.
• It repeats itself across batches. Run it weekly and you will see the same angles. Tell it what you already published.

The research half of this, working out what to make rather than making it, is on The growth guide. The short-video versions are on Reels and TikTok. New to Claude Code? Start here.