Press release June 3, 2026

Ideogram releases 4.0, a frontier image model with open weights, and a new brand for the company.

The release closes the quality gap between closed-source frontier image models and the open ecosystem, and lays the foundation for a layer-based generation stack designers, developers, and research teams can build on.

TORONTO, June 3, 2026. Ideogram today released Ideogram 4.0, a frontier text-to-image foundation model, as an open-weight release with a commercial license. The model closes the quality gap between proprietary frontier image models and the open ecosystem, and marks the start of a generation stack designed around how production design work is actually shipped: as layers, not as flat frames. Ideogram is also introducing a new brand identity for the company.

• Ideogram 4.0 is released with open weights under a commercial license, available for download today and through the Ideogram API.
• Enterprises can fine-tune the model on their own brand and product data and deploy it inside their own environment. The weights, the fine-tunes built on top of them, the training data, and every inference call stay on customer infrastructure.
• The model delivers frontier-grade text rendering across languages, bounding-box layout control, and 2K photoreal output.
• 4.0 is the foundation for a layer-based generation stack: transparent layers today, editable text and movable image layers in a follow-up release, and branded-asset generation with typography, palette, and logo fidelity soon after.

The gap this release closes

Over the last two years, the largest closed labs have moved the frontier of image generation forward across three axes: rendering text in scene at production fidelity, following detailed prompts accurately, and producing photographic and illustrative output that holds up next to directed photography. The open ecosystem has not kept pace. The gap between proprietary frontier image models and the strongest open-weight alternatives widened to the point that a real choice between them, for serious advertising, marketing, and enterprise production work, no longer existed.

Ideogram 4.0 closes that gap. It is the first model in Ideogram's frontier line released with open weights, and it brings the capabilities that previously lived only behind proprietary APIs into a release that customers can download, audit, fine-tune, and run on their own infrastructure.

A specialized foundation, not a unified multimodal model

A second narrative has shaped the field in parallel. As large multimodal models that combine text and image tokens in a single unified architecture have grown more capable, a view took hold that specialized image models were obsolete, and that text-to-image work would be subsumed into general-purpose multimodal systems.

Ideogram has bet the other way. The hardest problems at the forefront of design generation, headline-grade typography, deterministic layout, brand fidelity, layered output, do not come for free out of unified multimodal systems. They require a foundation engineered for those problems specifically. Ideogram 4.0 is that foundation, and the release shares the research insights behind it with the developer and academic community in service of further work on these problems.

What 4.0 ships today

Three capabilities anchor the release.

Text rendering at production fidelity. Ideogram has led on typography in scene since launch, and 4.0 extends that lead with multilingual support, denser type at smaller scales, and reliable rendering of headlines, packaging copy, and signage that say the right words.

Bounding-box layout control. Users can specify where a logo, headline, callout, or subject belongs on the canvas. Layout is no longer something the model samples and the designer corrects afterward. It is something the brief directs.

Photoreal output at 2K. The model produces images that hold up next to directed photography or a magazine spread, and supports fine-tuning so customers can converge it on a house style.

From flat frames to layer-based design

Most professional design work is not a single pixel layer. Headlines change after the model has done its job. Cutouts drop onto new backdrops. Branded assets follow a typography system and a color palette. Ideogram 4.0 is the start of a generation stack built around those realities, rolled out across this release and the next.

Today, transparent layers are available through the Ideogram Background Remover, which produces a clean alpha cutout from any 4.0 output. In the follow-up 4.0 release, editable text and movable image layers come directly out of inference, returning the model's output as a stack of components rather than a flat frame. Soon after, the model will generate branded assets that follow a customer's typography, color palette, and logo without manual cleanup.

Each of these capabilities is shipping or scheduled, not conceptual. The roadmap reflects a single bet: the foundation under design work is not one more flat image. It is a model that returns the pieces designers and developers can keep working with.

Open weights under a US-headquartered jurisdiction

Open-weight image models are no longer scarce, but the frontier-grade ones increasingly originate in jurisdictions that bring constraints into the weights and into the data path. Independent evaluations of DeepSeek-R1 have found it 11 times more likely to produce dangerous outputs and 4 times more likely to generate insecure code than Western alternatives, and that prompts touching Chinese political topics raise the rate of security-vulnerable code by roughly 50 percent. Australia, the United States Navy, and the United States House of Representatives have restricted Chinese-origin AI tools from government and military networks. For governments, defense contractors, and regulated enterprises, the choice of which open weights to standardize on is also a choice of jurisdiction.

Ideogram is headquartered in Toronto and San Francisco. Ideogram 4.0 ships under a commercial license with no embedded political alignment in the weights, no Chinese data-handling exposure, and a model card that documents training data and behavior in full. Customers can run the model on their own hardware, audit it end to end, and operate it under the residency and compliance rules their industry actually requires.

For developers and research teams

The audience for an open-weight frontier image model extends well past design teams. Robotics labs, autonomous-driving stacks, and world-model researchers use image generation as a component inside larger learned systems: synthetic perception data, domain randomization, counterfactual evaluation, and the visual backbone of frame-by-frame world models. None of this work is possible against a closed API; it requires gradients, intermediate features, and the freedom to fine-tune the model on a research team's own domain.

Ideogram 4.0 is built for that audience too. Bounding-box conditioning is native, so compositional control no longer depends on a ControlNet pipeline. Text rendering across languages makes synthetic data with readable signage, license plates, and storefronts possible without manual labeling, so a dataset can generalize from one geography to the next. The commercial license is sized to survive being embedded inside a shipped product.

A new brand for the company

Ideogram is also introducing a new brand identity alongside the model release. The new identity reflects the company's expanding scope: a research lab focused not only on text in images but on the foundation under modern design and visual-systems work. Brand materials will roll out across ideogram.ai, the Ideogram product, and developer surfaces over the coming weeks.

Many in the field believed that specialized image models were going to be subsumed by unified multimodal systems. Our view has been that the hardest problems at the forefront of design generation, headline-grade typography, deterministic layout, branded layered output, need a foundation engineered for them. Ideogram 4.0 is that foundation, and we are releasing it openly so developers, enterprises, and research teams can build on it.

Mohammad Norouzi, co-founder and CEO, Ideogram

Availability

Open weights are available on GitHub today. Hosted access is live on ideogram.ai and through the Ideogram API. Commercial license tiers are listed at ideogram.ai/licensing. Enterprise licensing inquiries can be directed to [email protected].

Media contact
[email protected]