live
MODEL IDideogram:4.6@0

Ideogram 4.6

Ideogram
by
Available with Zero Data Retention

Ideogram 4.6 is Ideogram's image generation and editing model, with native output up to 4096x4096. It generates images from text at any size on a 32-pixel grid, with an optional transparent background, and edits a source image using up to four more images as references. It keeps characters recognizable across scenes and matches exact colors more reliably than Ideogram 4.5, which suits product shots, brand recoloring, and campaign variations.

Ideogram 4.6

Text and design

How to use Ideogram 4.6 for designs where the words have to be right: quoted copy, type hierarchy, placement, dense layouts at 4K and non-Latin scripts.

Introduction

An image model that learns lettering as texture knows what text looks like, not how a word is spelled, so a poster comes back with a headline that almost reads. Ideogram builds its models around the opposite: text that is spelled right and placed where you asked.

Ideogram 4.6 carries that into event posters, packaging, app screens and printed menus, and at up to 4096 pixels on a side it renders small print at a size a printer can use. The poster below is one prompt, with no text added afterward.

This guide covers quoting exact copy, building a type hierarchy, placing text on a layout, dense layouts at 4K and non-Latin scripts. Every prompt here runs with Magic Prompt off, so the layout that comes back is the one the prompt describes.

Quoting exact copy

Put every word that must appear inside double quotes. The quotes mark the difference between copy to render and a description of the image, and they are how the model knows where one line ends and the next begins.

Try in Playground
import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'ideogram:4.6@0',
  positivePrompt: 'A clean mobile app subscription screen for a language learning app, shown flat and full-frame with no phone frame: a warm off-white background, a friendly flat illustration of a speech bubble and a small globe in the upper third, the headline "Speak Spanish in 10 minutes a day" in a bold rounded sans-serif below it, three short lines with green check marks reading "Daily lessons that fit your commute", "Practice with real conversations" and "Track your streak", a pale green card reading "$9.99 / month" with "Cancel anytime" under it in small gray type, and a wide dark green button reading "Start free trial" at the bottom.',
  width: 1152,
  height: 2048,
  settings: {
    magicPrompt: 'off'
  }
})
import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "ideogram:4.6@0",
            "positivePrompt": "A clean mobile app subscription screen for a language learning app, shown flat and full-frame with no phone frame: a warm off-white background, a friendly flat illustration of a speech bubble and a small globe in the upper third, the headline \"Speak Spanish in 10 minutes a day\" in a bold rounded sans-serif below it, three short lines with green check marks reading \"Daily lessons that fit your commute\", \"Practice with real conversations\" and \"Track your streak\", a pale green card reading \"$9.99 / month\" with \"Cancel anytime\" under it in small gray type, and a wide dark green button reading \"Start free trial\" at the bottom.",
            "width": 1152,
            "height": 2048,
            "settings": {
                "magicPrompt": "off"
            }
        })


asyncio.run(main())
curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "imageInference",
      "taskUUID": "c3a8f1e6-7d24-4b90-9e5a-2f6b8d1c4a73",
      "model": "ideogram:4.6@0",
      "positivePrompt": "A clean mobile app subscription screen for a language learning app, shown flat and full-frame with no phone frame: a warm off-white background, a friendly flat illustration of a speech bubble and a small globe in the upper third, the headline \"Speak Spanish in 10 minutes a day\" in a bold rounded sans-serif below it, three short lines with green check marks reading \"Daily lessons that fit your commute\", \"Practice with real conversations\" and \"Track your streak\", a pale green card reading \"$9.99 / month\" with \"Cancel anytime\" under it in small gray type, and a wide dark green button reading \"Start free trial\" at the bottom.",
      "width": 1152,
      "height": 2048,
      "settings": {
        "magicPrompt": "off"
      }
    }
  ]'
runware run ideogram:4.6@0 \
  positivePrompt="A clean mobile app subscription screen for a language learning app, shown flat and full-frame with no phone frame: a warm off-white background, a friendly flat illustration of a speech bubble and a small globe in the upper third, the headline \"Speak Spanish in 10 minutes a day\" in a bold rounded sans-serif below it, three short lines with green check marks reading \"Daily lessons that fit your commute\", \"Practice with real conversations\" and \"Track your streak\", a pale green card reading \"\$9.99 / month\" with \"Cancel anytime\" under it in small gray type, and a wide dark green button reading \"Start free trial\" at the bottom." \
  width=1152 \
  height=2048 \
  settings.magicPrompt=off
{
  "taskType": "imageInference",
  "taskUUID": "c3a8f1e6-7d24-4b90-9e5a-2f6b8d1c4a73",
  "model": "ideogram:4.6@0",
  "positivePrompt": "A clean mobile app subscription screen for a language learning app, shown flat and full-frame with no phone frame: a warm off-white background, a friendly flat illustration of a speech bubble and a small globe in the upper third, the headline \"Speak Spanish in 10 minutes a day\" in a bold rounded sans-serif below it, three short lines with green check marks reading \"Daily lessons that fit your commute\", \"Practice with real conversations\" and \"Track your streak\", a pale green card reading \"$9.99 / month\" with \"Cancel anytime\" under it in small gray type, and a wide dark green button reading \"Start free trial\" at the bottom.",
  "width": 1152,
  "height": 2048,
  "settings": {
    "magicPrompt": "off"
  }
}
Response
[
  {
    "taskType": "imageInference",
    "taskUUID": "c3a8f1e6-7d24-4b90-9e5a-2f6b8d1c4a73",
    "imageUUID": "7a2d5f9c-1e83-4c46-b0a7-9d3e6f1b2c58",
    "imageURL": "https://im.runware.ai/image/os/a14d18/ws/2/ii/7a2d5f9c-1e83-4c46-b0a7-9d3e6f1b2c58.jpg"
  }
]

Give every quoted string its own place and its own style in the prompt. Words outside the quotes are read as description, so "green check marks" came back as drawn check marks rather than as text.

The screen is 1152 × 2048, a 9:16 frame on the 32-pixel grid. Any size on that grid works, so generate at the shape of the screen the design is for instead of cropping a square.

Building a type hierarchy

A layout reads in order because its lines differ in size, weight and style. Describe that for each line, from top to bottom.

The bag has three levels, and the prompt gives each one a size, a style and a position. Give the largest size to the line a shopper reads first, here the recipe. "A clear type hierarchy" tells the model the sizes are meant to contrast.

Describe type by its traits: serif or sans-serif, weight, width, capitals or lowercase, letter spacing.

Placing text on the layout

For a layout with several blocks of copy, anchor each block to a region of the frame, such as the left half, a panel or a footer strip.

The prompt reads like a layout spec: the photo takes the left half, the navy panel the right, and inside the panel each line has a slot from top to bottom, down to the rule above the footer. Every block has a slot, which is what makes the prompt a template: change the name, the title and the photo to make the next speaker's card.

Dense layouts at 4K

A menu or a spec sheet packs many short strings into one frame, and at print size the small lines are what break first. Organize the prompt the way the layout is read: describe the shape of one item once, then list every item as labeled fields in reading order.

After stating the layout of a dish once, the prompt lists every dish as numbered, labeled fields, section by section. The labels keep each string attached to its dish, so the nine prices land on the right nine lines. This menu also ran at high, the tier for the version that ships.

At 2720 × 4096, the italic line at the foot of the page is real text with clean letterforms, sized to read on a printed page, and the dotted leaders stay crisp. A 2K render of the same 2:3 page holds about a third of those pixels, so render dense print layouts at the size they will be printed.

Keep each item to a name and a price. Names, prices and a short line of small print render reliably. A one-line description under each of nine dishes is where a list this long drifts, with descriptions landing on the wrong dish, so set descriptions and body copy in your own layout on top of the image.

Non-Latin scripts

Quoted copy can be in another script. Paste the text in its own script, and describe its direction and style the same way you would for English.

Write the direction into the prompt. The Arabic banner asked for a right-to-left layout and got one, and the discount kept the Western digits it was written with, so type digits in the form the design should print.

Have a native speaker check every non-Latin result before it ships. A wrong character in a script you do not read looks exactly like a right one.

Tips

  1. Quote every word that must appear. The quotes separate the copy from the description of the image.

  2. Give each line a size, a style and a place. Describe the hierarchy line by line, from top to bottom.

  3. Anchor blocks to regions. A half, a panel or a footer strip gives every piece of copy a slot, and a layout with fixed slots can be reused as a template.

  4. State a repeated pattern once. "The dish name on the left and its price aligned right, with a thin dotted leader between them" covers every item, and numbered, labeled fields keep each string on its line.

  5. Render print layouts at print size. Small type needs the pixels, and 4096 on a side gives them.

  6. Turn Magic Prompt off for layouts. The layout that comes back is then the one the prompt describes.

  7. Proofread before shipping. Check prices, dates and every non-Latin line, ideally with a native speaker.

To hold a design to exact brand colors, see Brand colors.