Aristotto
Back
Guides

How to Prompt Ideogram 4

Aristottoby Aristotto4 min
A golden retriever sitting on a porch, holding a wooden sign reading 'Adopt Me' in its mouth.

Ideogram 4 released June 3, 2026, and it's still the specialist pick when a piece of design work lives or dies on typography, a poster headline, packaging copy, a logotype. Where it does something genuinely different from most other image models: for tight style control, a structured JSON prompt outperforms a plain-language paragraph.

Why JSON, and in what order

Most models we cover reward natural sentences. Ideogram 4 is the exception for style-critical work. Field order matters here, and the order that holds up in practice is: aesthetics, then lighting, then photo details, then medium. Putting aesthetics first anchors the overall visual direction before the model starts filling in specifics, and medium last locks the final rendering treatment without it fighting earlier decisions.

A structure to work from:

{
  "aesthetics": "minimalist editorial, high contrast",
  "lighting": "single hard directional light, deep shadow",
  "photo": "close crop, centered subject, shallow depth of field",
  "medium": "35mm film photograph, grain"
}

Color: keep it to a handful of hex values

Specify two to three hex values per shot rather than a full swatch dump. Overloading the palette field doesn't give the model more control, it gives it more to reconcile, and the result usually muddles toward an average of everything you listed instead of committing to any of it.

Positioning: switch back to plain language

Here's the one place JSON stops helping. For where something sits in the frame, plain language reads far more reliably than bounding-box coordinates. "Centered in the frame with even space on both sides" works. A set of x/y coordinates does not translate the way you'd expect, the model interprets spatial language, not a layout grid.

Text and typography

This is where Ideogram 4 earns its place on the roster. Put exact copy in quotes, and if you're specifying a headline plus a subhead plus a small tagline, label each one so the model knows the hierarchy rather than guessing which line is the emphasis.

Three prompts that work

Poster with headline

{
  "aesthetics": "minimalist concert poster, bold and graphic",
  "lighting": "flat, no dramatic shadow, poster-style flat lighting",
  "photo": "centered composition, generous negative space around the edges",
  "medium": "screen-print poster, slight paper grain texture"
}

Headline text, large and centered: "NIGHT SESSIONS". Subhead beneath it, smaller: "Live at The Green Room, Sept 12". Color palette: #0B0C10, #66FCF1, #C5C6C7.

Product packaging label

{
  "aesthetics": "premium artisanal food packaging, clean and modern",
  "lighting": "soft even studio lighting",
  "photo": "flat label design, front-facing, no perspective distortion",
  "medium": "printed label mockup, matte paper texture"
}

Label text: brand name "MAPLE AND OAK" in bold serif at the top, product name "Small Batch Honey" beneath it in a lighter weight. Color palette: #2C1810, #D4A574, #FAF3E0.

Editorial cover typography

{
  "aesthetics": "high-fashion magazine cover, editorial and bold",
  "lighting": "dramatic single-source studio light",
  "photo": "tight crop, subject's face fills most of the frame, generous space reserved at the top for masthead text",
  "medium": "glossy magazine cover, photographic"
}

Masthead text at the very top, large and bold: "ATELIER". Cover line beneath it, smaller: "The Quiet Power Issue". Color palette: #1C1C1C, #FFFFFF, #B8975A.

Where this stands with actual designers

Ideogram 4 shipped as the company's first open-weight model, and the reception among working professionals is genuinely strong. ContraLabs ran a blind typography evaluation judged by ten professional designers pulled from Contra's top-earning talent pool. Ideogram 4 won first place 47.9% of the time, well ahead of Nano Banana 2 at 30.0%, FLUX.2 at 15.5%, and Grok Imagine 1.0 at 15.0%. Asked whether they'd actually use it in real client work, the same designers rated it highest at 3.55 out of 5, meaningfully above every model it was tested against.

On the independent design-focused arena that tracks general text-to-image preference, Ideogram 4 ranks ninth overall and first among every open-weight model on the board, an unusual result for a model that publishes its weights rather than keeping them closed.

One honest nuance: community reaction on release wasn't uniformly positive. Real discussion threads focused less on output quality and more on licensing terms and whether "open" was the right word for a release that isn't fully unrestricted for commercial self-hosting. Worth knowing if you're evaluating the self-hosted weights specifically rather than using it through a hosted platform.

Common questions

Does JSON formatting actually change the output, or is it just organization?

It changes the output. Field order shifts how the model weighs aesthetic direction against technical detail, this isn't just about readability.

Why does bounding-box positioning give inconsistent results?

Ideogram 4 interprets spatial language rather than coordinate systems. Describe position the way you'd describe it to a person.

How many colors should I specify?

Two to three hex values per shot. More than that dilutes rather than sharpens the result.

Discover more

View all