Make a product image and put it on a live site

Make a hero image for PRODUCT_NAME and put it on a one-page site.
Product: PRODUCT_NAME, PRODUCT_ONE_LINE. Audience: AUDIENCE.

1. Call image_generate once to make the hero image: STYLE_AND_SUBJECT. It is for a landing-page hero, wide (16:9). The image must contain no words, letters, or logos. Save it with output_dir "site/images" and file_stem "hero".
2. Open the file with read. Tell me what the image shows. If it has any letters, or it does not match the brief, make at most one replacement with a corrected prompt; two images in total is the limit. Say which one you kept.
3. Write /site/index.html: one page with the headline PRODUCT_HEADLINE and one sentence PRODUCT_SENTENCE as HTML text (not inside the image), the image with a useful alt text, and no external scripts, fonts, or analytics.
4. Check that every file the page references exists in /site and that the page loads from a local static server without errors. Then call plori_publish_ready for /site as a static site. Do not publish it yourself.
5. Reply with the image prompt you used, the file paths, how many images you generated, and the cost of the generation.

Why is the prompt shaped this way?

Each image costs credits, so the prompt sets the number of images before the run starts: one image, and at most one replacement.

The prompt puts the headline and the sentence in the HTML. This guide has no test of letters inside a generated image. HTML text shows the exact words that you type. Screen readers can read it, and you can change it without a new image.

The image goes in site/images, next to the page. The /site directory then holds every file that the page needs. The tool adds a number to the file name, and it selects the extension from the image format. With file_stem "hero", the file in this run was hero-1.jpg.

Step 2 tells the agent to open the image and say what it shows. Not every model can see an image, so look at the image yourself.

Step 4 makes the agent test the page before it gives the page to you. Then the agent calls plori_publish_ready, which shows a Ready to publish card. The agent cannot publish the site.

Where did the run start, and what needed approval?

The recorded run started from the plori CLI, signed in with an API key. This sign-in uses the same approval rules as the Plori web app, so the web app runs this prompt in the same way. The image tool is not available on runs that start over the Plori MCP connector or from a CLI session signed in through the browser.

No call needed approval. Image generation does not ask for approval, and the agent wrote files only in its Workspace. The only action outside the Workspace is publishing, and you do it in the web app.

How do you publish the page?

Publishing puts the page on the internet at https://<name>.plori.site. Only you can publish it.

1. Open the agent conversation in the Plori web app.
2. On the Ready to publish card, click Review & publish site.
3. Make sure that the directory is /site. The dialog finds index.html and shows the directory as a static site.
4. Keep the suggested name, or type a different name for the address. Use lowercase letters, digits and hyphens, with a minimum of 3 characters.
5. Select Private or Public. Private is the default.
6. Click Publish privately or Publish publicly.
7. Open the site. For a private site, click Open site in the Files preview. Open site gives your browser access to the private address.

If the card is not in the conversation, use Files. Open site/index.html and click Publish as a website. Each agent has one plori.site address. The name must be unique on Plori, and it stays reserved to the agent after you unpublish the site.

What did the real run prove?

On 17 October 2026, we gave the prompt these values:

• Product: Plori, "a team of AI agents that collaborate on one product in the cloud".
• Audience: founders and small product teams.
• Style and subject: "an abstract dotted sphere of light in the dark with small bright points around it and thin arcs of light between them, one warm orange accent, calm and clean, no people".
• Headline: "A team of AI agents collaborates on one product."
• Sentence: "Code, research, content: each works on its own copy, and Plori merges their changes back."

The run started at 21:00:21 UTC and completed in 1 minute 44 seconds. The agent made one image_generate call. It wrote this image prompt from the brief (shortened here):

Abstract hero image for a SaaS landing page: a large sphere suspended in deep near-black darkness, formed entirely of thousands of tiny glowing dots of pale cool white-blue light ... Wide 16:9 composition with the sphere slightly off-center and open dark space on the left for headline text.

The tool reported one image, site/images/hero-1.jpg: a JPEG of 139 KB and 1376 by 768 pixels, at aspect ratio 16:9 and the default resolution of the model. That model has one resolution, 1K. This is the image from the run:

We looked at the file. The sphere is right of center, and the left part of the image is dark and empty. Thin curved lines go around the sphere, with small bright points on them. One orange point is at the upper right. We found no letters in the image.

The agent opened the file with read and kept the first image. The prompt told it to say what the image shows. Its reply did not describe the image. It said only that the image matched the brief. This guide does not use that statement as evidence.

The agent wrote /site/index.html (5,853 bytes). The page shows the headline and the sentence as HTML text on top of the image, and the image has alt text. The page has no script tags and loads no external fonts or analytics. The page also has four lines of text that the prompt did not ask for. They are a line above the headline, a note in the header, a line about the audience and a sentence in the footer.

Before the handoff, the agent started a local static server. The page and the image both returned HTTP 200, with the types text/html and image/jpeg. The agent found no missing file and no external resource. Then it called plori_publish_ready for /site as a static site, with Private as the default. It did not publish the site. The Workspace saved both files when the run ended.

On 17 October 2026 the owner published /site as a public static site at https://plori-hero.plori.site/. The publish call went through the site API with the owner's API key. It is the same action as the publish dialog. A GET request for the page returned 200 text/html, and a GET request for /images/hero-1.jpg returned 200 image/jpeg with 139,363 bytes. A HEAD request returned 403, so check a site with GET.

The run cost $0.061 in credits, from the usage records of the run:

Item	Cost	Detail
Image	$0.034	One image. The tool reported $0.0336.
Run time	$0.017	1 minute 44 seconds
Model usage	$0.010	11 model calls
Total	$0.061

The run proves that this prompt produces an image, a page for the image and a publish card. The agent took no action outside the Workspace. The run does not prove that the agent checked the image. Check the image and the page text yourself before you publish.

How do you check it safely?

The agent cannot publish the site. The image costs credits, and the prompt limits the run to two images. Test the page as a private site first:

1. Run the prompt.
2. Open site/images in Files. Make sure that the directory has one or two images.
3. Look at the image. Make sure that it has no letters and that it agrees with your brief.
4. Open site/index.html. Read all text on the page.
5. Ask the agent to remove the text that you do not want.
6. In site/index.html, read the alt text of the image. Make sure that it describes the image.
7. Publish the site as Private.
8. Click Open site. Make sure that the page shows the image and the text.
9. If the page must be public, open Site settings. Change the visibility to Public.
10. Open the public address in a browser where you are not signed in to Plori.

To stop the site, open Site settings and click Unpublish. The name stays reserved to the agent.

What are the limits?

• The image model has one resolution, 1K. The model selects the pixel size. In this run, a 16:9 image was 1376 by 768 pixels. The tool does not make an image of an exact size such as 1200 by 630.
• Each call makes one image. The prompt allows two calls.
• The model accepts 14 aspect ratios: 1:1, 1:4, 1:8, 2:3, 3:2, 3:4, 4:1, 4:3, 4:5, 5:4, 8:1, 9:16, 16:9 and 21:9. The tool replaces a ratio that is not in the list with one that is.
• Letters inside a generated image can be wrong, and this guide has no test of them. Keep all text in the HTML.
• The agent might not see the image. That depends on the model that runs the turn. In this run, the reply did not describe the image. You must look at it yourself.
• The agent can add text that you did not ask for. In this run, it added four short lines. Read the page before you publish it.
• Do not use image generation for charts, diagrams or exact labels. Use code for them, for example SVG.
• Each call is independent, so two images can have different styles. The tool accepts reference images of 1 MiB or less, but this guide does not test them.
• A static site cannot run a server. For a form, an API or a secret, use a service. The scale-to-zero service guide shows how.
• A static site shows the files as they were when you published. To show later changes, publish again.
• Image generation is available in the Plori web app and in the plori CLI signed in with an API key. Runs that start over the Plori MCP connector do not have it.

Common questions about product images with an agent

Can an AI agent make an image for my landing page? Yes. In this run, a Plori agent generated one 16:9 image, wrote a one-page site for it, and tested the page. Plori runs AI agents that collaborate on one product in the cloud, and image generation is one of the built-in tools.

How much does one image cost? In this run, one image cost $0.034. The tool reported $0.0336. The whole run cost $0.061, including run time and model usage.

Can the agent publish the site without me? No. The agent shows a Ready to publish card. You choose the address and the visibility, and then you publish the site.

Can I run this from Claude Code or another MCP client? Not this job. Runs that start over the Plori MCP connector do not have the image tool. Use the Plori web app, or the plori CLI signed in with an API key.

Can I put my headline inside the image? This guide does not test text inside images. Put the headline in the HTML.

Can I get an image of an exact size, such as 1200 by 630? No. You can ask for an aspect ratio. The model selects the pixel size. In this run, a 16:9 image was 1376 by 768 pixels.

How do I change the page after I publish it? Ask the agent to change the files. Then open Publish on the site directory in Files. Select Save.

Related

• Publish a site, for the publish dialog, visibility and address names
• Run a Node web service that scales to zero, for a site that needs a server
• plori CLI, for API key sign-in
• What a credit buys, for the balance and the usage ledger