Generating Images Right in the Chat

Ask Mythex for a hero image, an og:image or a logo mark and it draws one in the chat, then puts it in your app only when you say so. Here's how it works.

Mythex Team · 2026-09-29 · 4 min read

You can ask Mythex for a picture in plain language — a hero image, a background, an og:image with your words on it, a logo mark — and it draws one right in the chat. It stays in the chat whether or not it ever goes into your app, and it only goes into your app when you ask. Two recent changes shaped how this works: the agent now generates an image only when you ask for one, one per message, and every file you attach, including the photo you want edited, now reaches the agent.

What it does

In a project's chat, say what you want:

"generate an image of a mountain lake at sunrise"
"make a hero image for the landing page: a ceramic mug on a wooden desk"
"I need an og:image with the words Ship Faster on it"

Or use the Create image command: type /image, or pick it from the + menu next to the composer, then describe the picture.

While it's being drawn you see a placeholder in the image's final shape, and the finished picture replaces it exactly where it sits, so nothing jumps around. Click any generated image to open it full size.

Your app doesn't change until you say so

Generated images are saved to .mythex/images/ in your project, separate from your app's own source. Asking for a picture doesn't touch your app.

When you want one in it, say so: "put that on the homepage", "use it as the hero image", "add it to the About section". Mythex copies it to wherever your project serves static files from — public/ for Vite, Next or Remix, static/ for Django or SvelteKit, and so on — and wires it into the page. It works that out from your project rather than assuming, so the same request works whatever you built with.

Editing a picture you already have

You don't have to start over to change something:

"take that trail sign image and carve 'Hello World' into it"
"make the lighthouse image a daytime scene"
attach your own photo and ask for an edit

The original stays as it was; the edit is saved as a new image.

Attaching your own photo got more reliable in the same round of work. Before, some attachments never reached the agent: a PDF sent together with a screenshot could be dropped, and iPhone HEIC photos or large images could quietly vanish. Now every file you attach gets its own entry the agent can see — described, converted to text, or named for what it is with a reason if it can't be read. You can attach up to 20 files per message, images up to 20 MB and other files up to 50 MB.

Why only when you ask, and one at a time

Our first version let the agent generate up to several images per message on its own judgement. In practice that went wrong in two ways. The agent illustrated a birthday site nobody had asked it to illustrate. And when a message was sent twice, a full set of scene images was bought twice.

Each generated image costs you credits, so the decision should be yours. The rules now are:

The agent only generates when you ask — in your message, or through the Create image command.
Pictures you didn't ask for come from stock. If a site needs ordinary photography — a person at a laptop, a plate of food — the agent searches stock photo libraries instead. That's free and instant, and usually better than a generated substitute. If it thinks a generated image would clearly be better, it says so in one line and lets you ask.
One generated image per message. If you ask for several, it makes the one that matters most, finishes the work with it, and tells you that you can ask for the next. This is enforced in the tool itself, not just suggested to the agent.
A failure doesn't use your turn. If a generation fails, the slot is handed back and you aren't charged for it.
Repeats are reused. Before generating, the agent checks .mythex/images/. If a message repeats an earlier request, the image may already be there, and reusing it costs nothing.

Generate or search?

Generation is for what stock photos can't give you:

art that matches your app's own palette or style
a specific scene or composition ("the product box on a workbench, top-down")
an illustration, pattern, texture or abstract background
an og:image or social card with your product's own words on it
a logo or app mark

For logos, Mythex generates the symbol and sets the name as real text beside it. Generated lettering is unreliable at small sizes, and live text is something you can restyle later.

Shape

Shape comes from what you ask for, or from where the image is going: a hero, banner or og:image comes back wide, an avatar or tile comes back square, and a phone splash comes back tall. You can also just say "16:9", "1:1" or "9:16".

Limits

One generated image per message.
Output is always JPEG.
Logos come back as a drawn mark plus real text, not generated lettering.
Each generated image costs credits; a stock photo search costs nothing. See Credits.

The docs have the details: Generate images. For making the rest of your app look as good as the pictures, see how to design a good-looking app with AI.