Ideogram 4.5

Introduction

Ideogram 4.5 generates an image from a text prompt of up to 10,000 characters, at one of 36 fixed sizes that run from 1K to 2K. The prompt carries almost all of the control. Around it sit three settings: Magic Prompt, which can rewrite the prompt before generation, the size, and a quality tier.

Advertising photograph for a sportswear brand: a woman in her early thirties in a charcoal running jacket and black tights crouches to tie the lace of a white running shoe on a stone waterfront promenade at sunrise, with the city skyline soft and hazy across the water behind her. She sits in the right third of the frame. Warm low sunlight from the left, long shadows, a light mist over the water. Shot on a 35mm lens at knee height, shallow depth of field, crisp detail on the shoe and laces, natural color grading.

This guide covers the order that makes a prompt work, planning empty space for copy, what Magic Prompt does to short and long prompts, choosing a size and choosing a quality tier.

Building the prompt

Ideogram 4.5 follows a long, specific prompt closely, so the work is in the order of the details. The prompts in this guide all follow the same one: what the image is for, then the subject, the setting, where things sit, the light and the camera.

Commercial product photograph for a 16:9 web banner: a matte sage green insulated water bottle with a bamboo cap stands on a flat wet granite rock beside a clear mountain stream. The bottle sits in the right third of the frame, and the left half is calm, out-of-focus pine forest with room for copy. Morning light from the left, a soft rim light on the bottle, fine water droplets on its surface. Shot on a 100mm lens at eye level, shallow depth of field.

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:4.5@0',
  positivePrompt: 'Commercial product photograph for a 16:9 web banner: a matte sage green insulated water bottle with a bamboo cap stands on a flat wet granite rock beside a clear mountain stream. The bottle sits in the right third of the frame, and the left half is calm, out-of-focus pine forest with room for copy. Morning light from the left, a soft rim light on the bottle, fine water droplets on its surface. Shot on a 100mm lens at eye level, shallow depth of field.',
  width: 2560,
  height: 1440
})

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:4.5@0",
            "positivePrompt": "Commercial product photograph for a 16:9 web banner: a matte sage green insulated water bottle with a bamboo cap stands on a flat wet granite rock beside a clear mountain stream. The bottle sits in the right third of the frame, and the left half is calm, out-of-focus pine forest with room for copy. Morning light from the left, a soft rim light on the bottle, fine water droplets on its surface. Shot on a 100mm lens at eye level, shallow depth of field.",
            "width": 2560,
            "height": 1440
        })


asyncio.run(main())

curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "imageInference",
      "taskUUID": "58c3e1a7-4b92-4d6f-8a05-e2f7c9b1d364",
      "model": "ideogram:4.5@0",
      "positivePrompt": "Commercial product photograph for a 16:9 web banner: a matte sage green insulated water bottle with a bamboo cap stands on a flat wet granite rock beside a clear mountain stream. The bottle sits in the right third of the frame, and the left half is calm, out-of-focus pine forest with room for copy. Morning light from the left, a soft rim light on the bottle, fine water droplets on its surface. Shot on a 100mm lens at eye level, shallow depth of field.",
      "width": 2560,
      "height": 1440
    }
  ]'

runware run ideogram:4.5@0 \
  positivePrompt="Commercial product photograph for a 16:9 web banner: a matte sage green insulated water bottle with a bamboo cap stands on a flat wet granite rock beside a clear mountain stream. The bottle sits in the right third of the frame, and the left half is calm, out-of-focus pine forest with room for copy. Morning light from the left, a soft rim light on the bottle, fine water droplets on its surface. Shot on a 100mm lens at eye level, shallow depth of field." \
  width=2560 \
  height=1440

{
  "taskType": "imageInference",
  "taskUUID": "58c3e1a7-4b92-4d6f-8a05-e2f7c9b1d364",
  "model": "ideogram:4.5@0",
  "positivePrompt": "Commercial product photograph for a 16:9 web banner: a matte sage green insulated water bottle with a bamboo cap stands on a flat wet granite rock beside a clear mountain stream. The bottle sits in the right third of the frame, and the left half is calm, out-of-focus pine forest with room for copy. Morning light from the left, a soft rim light on the bottle, fine water droplets on its surface. Shot on a 100mm lens at eye level, shallow depth of field.",
  "width": 2560,
  "height": 1440
}

[
  {
    "taskType": "imageInference",
    "taskUUID": "58c3e1a7-4b92-4d6f-8a05-e2f7c9b1d364",
    "imageUUID": "d1a7f4c9-3e26-4b85-9f0a-6c2e8b5d1f73",
    "imageURL": "https://im.runware.ai/image/os/a14d18/ws/2/ii/d1a7f4c9-3e26-4b85-9f0a-6c2e8b5d1f73.jpg"
  }
]

Lead with the format. "Commercial product photograph for a 16:9 web banner" tells the model what the image is for before it reads a single detail, and the composition decisions follow from that. The bottle sits in the right third and the left half stays soft because the prompt asked for both, in a sentence of its own.

The camera line comes last and does the photographic work: a 100mm lens compresses the background and a shallow depth of field keeps the forest from competing with the product.

Planning space for copy

Most commercial images get text laid over them later. If the prompt does not say where that text will go, the model composes for the subject alone, and whatever empty space is left ends up wherever it happened to fall.

Product photograph of a white ceramic reed diffuser with thin black reeds on a linen-draped side table, a small sprig of eucalyptus beside it, soft afternoon light.

Product photograph for a 4:5 social media post: a white ceramic reed diffuser with thin black reeds on a linen-draped side table, a small sprig of eucalyptus beside it, soft afternoon light. The diffuser sits in the lower third of the frame, and the top 40% of the frame is a calm, plain warm white wall with nothing on it, left empty for a headline.

The second prompt says three things the first one does not: where the product sits, how much of the frame stays empty, and what the empty part looks like. "A calm, plain warm white wall with nothing on it" matters as much as the 40%, because it describes the empty part instead of leaving it to the model.

Magic Prompt

settings.magicPrompt lets Ideogram rewrite your prompt before generating, expanding a short description into a fuller one. It takes auto, on or off, and auto is the default. The three images below come from the same seven-word prompt and the same seed.

A boutique hotel room with a sea view

A boutique hotel room with a sea view

A boutique hotel room with a sea view

autoonoff

All three came back as finished listing photos. Even with off, Ideogram 4.5 fills in the styling, the light and the camera position on its own, so a short prompt does not need rewriting to look done. What the setting changes is which choices get made: from the same seed, each one produced a different room.

A long prompt turns the question around. When every detail is already written, rewriting has nothing to add but its own choices.

Photorealistic hotel listing photograph of a bright boutique hotel room, shot from the doorway with a 24mm lens at chest height. A low double bed with crisp white sheets and a folded pale blue linen throw sits against the left wall under two small framed botanical prints. A natural rattan armchair stands in the right corner beside open glass doors that lead onto a small balcony with a view of a calm turquoise sea. Terracotta tile floor, whitewashed walls, soft morning daylight, no people.

Photorealistic hotel listing photograph of a bright boutique hotel room, shot from the doorway with a 24mm lens at chest height. A low double bed with crisp white sheets and a folded pale blue linen throw sits against the left wall under two small framed botanical prints. A natural rattan armchair stands in the right corner beside open glass doors that lead onto a small balcony with a view of a calm turquoise sea. Terracotta tile floor, whitewashed walls, soft morning daylight, no people.

Both follow the brief: the bed on the left under two prints, the rattan chair by the open doors, the sea beyond the balcony. The on version also added things nobody asked for, a potted olive tree on the balcony and a jute rug by the bed. With off, the model gets the prompt exactly as written, which is what you want when the prompt is a brief that a client signed off on.

Magic Prompt only applies to text-to-image. A request that includes inputs.referenceImages rejects the setting, since edits prepare their own instructions. See Editing with reference images.

Choosing a size

width and height must be one of the 36 preset pairs, and leaving both out lets the model pick one. The presets come in two tiers:

• 2K sizes such as 2048 × 2048, 2560 × 1440, 1792 × 2240 and 1440 × 2560, for delivery.
• 1K sizes such as 1024 × 1024, 1280 × 720, 896 × 1120 and 720 × 1280, for drafts and thumbnails.

The ratios run from square to 3:1 in either direction (3072 × 1024 and 1024 × 3072), which covers website heroes, stories, feed posts and display ad formats without cropping. The full list of pairs is on the model's API reference.

Wide real estate website hero photograph of a modern single-story lakeside house with floor-to-ceiling windows and a flat timber roof, seen from across a calm lake at dusk. Warm light glows from every window and reflects in the still water, and the deep blue sky fades to amber at the horizon. The house sits in the center, with lake and pine shoreline stretching to both edges of the frame. Shot on a 24mm lens, crisp architectural detail.

Vertical street style fashion photograph for a 9:16 story: a woman in her late twenties in a camel trench coat, white shirt and wide-leg black trousers crosses a sunlit city crosswalk mid-stride, looking ahead, a leather tote on her shoulder. Full figure from head to toe, the crosswalk stripes leading into the frame, soft blurred storefronts behind her. Late afternoon sun, shot on a 50mm lens at waist height.

Overhead food delivery app photograph for a 4:5 feed post: a salmon poke bowl in a matte white ceramic bowl with sushi rice, diced raw salmon, sliced avocado, edamame, cucumber ribbons, pickled red onion and a sprinkle of sesame seeds, a pair of wooden chopsticks resting on the rim. Pale concrete table, soft daylight from the top left, crisp detail.

Square e-commerce product photograph of a pair of white wireless earbuds in their open matte white charging case, resting on a smooth light gray concrete block against a soft gray background. A thin line of green light shows on the case. Soft studio light from above, a gentle shadow, centered, crisp detail.

Name the format in the prompt as well as in the size. "Wide real estate website hero" and "vertical street style photograph for a 9:16 story" tell the model how to compose for the frame it is filling, so the house spreads across the banner and the woman fills the story from head to toe.

The two tiers differ in how much detail the pixels can hold. The same prompt and seed at 1024 × 1024 and 2048 × 2048:

Close-up product photograph of a folded chunky cable-knit wool sweater in oatmeal, lying on a light oak table, the cable pattern and individual wool fibers sharply detailed. Soft window light from the left, shallow depth of field, e-commerce detail shot.

Close-up product photograph of a folded chunky cable-knit wool sweater in oatmeal, lying on a light oak table, the cable pattern and individual wool fibers sharply detailed. Soft window light from the left, shallow depth of field, e-commerce detail shot.

At the same magnification, the 2K inset resolves individual fibers where the 1K inset shows the knit as texture. Draft at 1K while the prompt is still moving, and switch to the 2K preset with the same aspect ratio for the version that ships.

Choosing a quality tier

settings.quality takes low, medium or high for text-to-image, and high is the default. The lower tiers return faster and cost less. The three results below share a prompt and a seed.

Commercial flat lay photograph on a pale pink marble surface, shot from directly above: a frosted glass pump bottle, a round white cosmetic jar and a squeeze tube arranged in a loose diagonal with a few sprigs of dried lavender. Each product carries a minimal label with the brand name "AURA" in thin black capitals and two lines of small print below it. Soft diffused daylight, gentle shadows, crisp detail.

Commercial flat lay photograph on a pale pink marble surface, shot from directly above: a frosted glass pump bottle, a round white cosmetic jar and a squeeze tube arranged in a loose diagonal with a few sprigs of dried lavender. Each product carries a minimal label with the brand name "AURA" in thin black capitals and two lines of small print below it. Soft diffused daylight, gentle shadows, crisp detail.

Commercial flat lay photograph on a pale pink marble surface, shot from directly above: a frosted glass pump bottle, a round white cosmetic jar and a squeeze tube arranged in a loose diagonal with a few sprigs of dried lavender. Each product carries a minimal label with the brand name "AURA" in thin black capitals and two lines of small print below it. Soft diffused daylight, gentle shadows, crisp detail.

lowmediumhigh

The arrangement holds across all three, and even low renders the brand name and the small print cleanly. The tiers differ more in the details the prompt left open, such as the product names on the labels, than in sharpness. Use low to check composition and wording while the prompt is still changing, and high for the version that ships.

Tips

1. Lead with the deliverable. "Product photograph for a 16:9 web banner" sets up every composition decision that follows.

2. Place the subject and the empty space. Say which third the subject sits in, how much of the frame stays empty and what the empty part looks like.

3. Turn Magic Prompt off for a finished brief. Rewriting adds details the brief never asked for.

4. Pick the size from the placement. Choose the preset that matches where the image runs, and name that format in the prompt too.

5. Draft at 1K and low, ship at 2K and high. Settle the prompt cheaply, then render the final at full size and quality.

6. Put exact copy in quotes. Text rendering has its own guide: Text and design.