How to Create Images in an AI Chat (Step by Step)
A hands-on guide to image generation right inside your chat. Turn on Create Image, pick a model, set aspect ratio and quality, and write prompts that work.
Here is the fun part: you do not need a separate app to make pictures. In MultiChats you can turn the same chat window into an image studio, type what you want, and watch it render in the thread. This guide shows you exactly how to create images in an AI chat, which of the five image models to reach for, how the aspect ratio and quality controls work, and how to write a prompt that actually gives you the picture in your head. It is built for total beginners, so no setup or art background needed.
Step by step: generate your first image
Image generation lives next to your normal chat, so you flip it on, type a prompt, and send. Here is the whole loop, with the web and mobile labels side by side.
Turn on image mode. On the web, click the
Create Imagebutton in the composer. On mobile, tap the+menu and chooseCreate image. An activeImagepill shows you it is on.Pick a model. Open the image config (the sliders icon) and choose from the five image models, grouped under OpenAI and Google. The next section breaks down which is which.
Set the shape. In the same config, pick an Aspect Ratio (square for avatars, wide for banners, tall for phone wallpapers). If you are on an OpenAI model, you also get a Quality control.
Write your prompt and send. Describe the subject, the style, and the mood in plain words, then hit send like any other message.
Iterate in the thread. Did not love it? Reply with a tweak ("make it warmer, lose the text") and send again. The chat keeps your earlier turns, so you can refine without starting over.
If this is your first time in the app, the getting started with multi-model AI guide walks through signing in and finding the composer, then come back here.
Which image model to pick
You get five image models, and they are not interchangeable. OpenAI's GPT ImageGen pair is strong at following instructions and rendering legible text in the picture. Google's Nano Banana line is fast and great for quick visual ideas, with the Pro tier aimed at the most detailed work. Here is the short version.
Model | Reach for it when |
|---|---|
| You want tight prompt-following and clean text in the image. A solid default. |
| You like the OpenAI look but want a lighter option for quick passes. |
| You want fast, fun first drafts to explore an idea. |
| You want the newer Google look with more detail than the original. |
| You want the most detailed, polished result for the final keeper. |
My honest habit: I start a concept on Google Nano Banana because it is quick to riff with, then switch to GPT ImageGen 2 or Google Nano Banana Pro once I know the composition I want. Changing the image model is the same idea as switching AI models mid-chat for text: you just reopen the picker and choose another one.
Aspect ratio, quality, and a prompt that lands
Two small controls do a lot of the heavy lifting before you even touch the words.
Aspect Ratio sets the frame. Square works for profile pictures and icons, wide (landscape) for headers and slides, tall (portrait) for posters and phone screens. Pick the shape your picture will live in before you generate, not after.
Quality (on the OpenAI models) trades speed for finish. A faster, lower setting is great while you are still figuring out the idea; bump it up for the final keeper.
Now the prompt. A good image prompt names four things: the subject, the style, the setting, and the mood. "A cat" is a coin toss. "A ginger cat curled on a windowsill at golden hour, soft film-photo style, warm and cozy" gives the model something to work with. A few rules of thumb that have saved me a lot of regenerations:
Lead with the subject, then layer on style and lighting. Front-load what matters most.
Say what you do not want. "No text, no watermark" is a common one.
If you need words in the image (a poster, a logo mock), lean on a GPT ImageGen model and spell out the exact text in quotes.
Refine in replies. Change one thing at a time so you can tell what actually moved the result.
A worked example, from vague to sharp
Say you want a logo-style badge for a coffee cart. Start vague and the model guesses everything: "a coffee logo" might land on a generic bean with random text. Tighten it and the result snaps into focus. Try "a minimalist circular badge logo for a coffee cart, the words ROAST AND ROLL in a bold sans-serif, warm cream and deep brown, flat vector style, no gradients, no extra text." That names the subject (badge logo), the style (flat vector, minimalist), the setting (coffee cart), the mood (warm), and the exact words you want rendered. Because there is legible text in it, a GPT ImageGen model is the safer bet here than the Nano Banana line.
What it costs (free, Pro, and Pro+)
This is where the web and mobile paths differ, so read carefully. On the web, free users get 2 image generations every 30 days. Use those up and you will see an upgrade prompt. On mobile, image generation needs a paid plan to even turn on, so there is no free allowance there.
A paid plan opens image generation up across the app. Pro is $20.99 a month or $199.99 a year, and Pro+ is $34.99 a month or $349.99 a year. One thing to know either way: each request returns a single image, so when you want variations, send the prompt again with a tweak rather than asking for four at once. The same subscription also opens up the rest of the model lineup, which is the whole point of having every major model in one app.
Quick image checklist
Run down this list before you hit send and you will waste far fewer generations:
Image mode is on (you can see the Create Image / Image pill active).
You picked a model that fits the job (text in the image? go GPT ImageGen).
Aspect ratio matches where the image will live.
Quality is low for drafts, high for the final pick (OpenAI models).
Your prompt names the subject, style, setting, and mood, plus anything to avoid.
FAQ
How do I generate an image in a chat app?
Turn on image mode, then type and send. In MultiChats, click Create Image on the web or pick Create image from the + menu on mobile, choose a model, describe what you want, and send. The picture renders right in the thread, so you can reply with tweaks to refine it. There is more background in our explainer on image generation in an AI chat.
Which image model should I use?
If your image needs readable text or you want tight prompt-following, start with GPT ImageGen 2. For fast idea exploration, Google Nano Banana is a great riff machine. When you want the most detailed, finished result, step up to Google Nano Banana Pro. There is no single right answer; switching between them mid-thread is part of the fun.
How many free image generations do I get?
On the web, free users get 2 image generations every 30 days. On mobile, image generation requires a paid plan, so there is no free allowance on phones. A failed generation does not burn your quota, only a saved image counts. To keep generating past that cap, or to generate on mobile at all, you need a Pro or Pro+ plan.
What is Google Nano Banana?
Google Nano Banana is the name of one of Google's image models in the picker, alongside Google Nano Banana 2 and Google Nano Banana Pro. The original is quick and playful for first drafts; the 2 and Pro versions push more detail. They sit next to OpenAI's GPT ImageGen models, so you can A/B the two families on the same prompt and keep whichever you like best.
Start making pictures
That is the whole workflow: flip on Create Image, pick a model, set the shape, write a clear prompt, and refine in replies. Free users on the web get two generations every 30 days to try it; if you want to keep generating past that, or make images on mobile, that is what the paid plans are for. Have a look at the plans and pricing and go make something good.