Your AI writer can make its own images
Posts go out without a picture because the thing writing them cannot see. Here is how to fix that.
3 min read · last checked
Why the posts have no picture
An agent that writes for you works in text. It can research, draft and publish, and at the point where a picture should go it has nothing to offer. So the post goes out bare, or with the same stock photograph as the last four, or with a placeholder that somebody means to replace and never does.
That is not a failure of the writing. It is a missing hand. The model doing the writing has no way to make a picture, and the tools around it were never connected to one.
Give it a generator it can call
This studio speaks MCP, the protocol that lets an assistant use a tool directly. Connect it once and your agent gets a set of functions it can call while it works: make an image, check whether it is done, fetch the file.
From the agent side it is three steps. Call generate_image with the prompt and, if you want to keep a look, the ids of reference images from your own library. Poll get_execution_status until it reports done. Fetch the file and publish it with the post.
- generate_image, makes the picture and returns an execution id.
- get_execution_status, tells you whether it is finished and hands you the file.
- list_assets, everything you have made before, so a series can keep one look.
- list_models, what each model is for and what kind of input it takes.
Prompt and images are one input, not two calls. Send the description and the reference pictures together and the studio picks the ingress that accepts both.
A cover that belongs to the article
The failure mode here is not "no picture" but "any picture". A generic office desk on top of a piece about pricing is worse than nothing: it tells the reader that no one was paying attention.
So derive the prompt from the piece itself. Take the one thing the article is actually about and photograph that thing. Our own guides do it: the one about lenses shows a row of bottles with the front one sharp, the one about keeping a character shows the same man three times in one room. Each cover is an example of what its guide teaches.
Give every image alt text, and write it from the same brief. An agent that can make a picture can describe it, and a post without alt text is a post some of your readers cannot use.
What it costs, and what it does not do
One cover is one generation. The exact amount depends on the model and the resolution, and it is on the button before anything runs, so a hundred posts a month is a number you can work out in advance rather than discover on an invoice.
What this does not do is decide. The studio makes what you ask for; it does not judge whether the picture suits your brand, and it will happily make forty of them. Keep a person on the last step, especially while you are still tuning the prompt.
Doing it on a schedule
The studio does the picture. It does not run your calendar and it does not post to your channels, and any tool that claims otherwise is selling you two products as one.
The scheduling belongs to whatever already runs your automation: a cron job, an n8n workflow, an agent platform with recurring tasks. That thing wakes up, asks your writer for the piece, asks this studio for the cover, and hands both to your CMS. We are one call in that chain, which is exactly what you want: replace any other link and this one keeps working.