Ideogram 4

Ideogram's first open-weight image model: 9.3B parameters, strong text rendering, JSON prompts and native 2K output. Non-commercial license.

In short. Ideogram 4 is an image model from the company behind the Ideogram image service. It is built to put exact, readable text into posters, logos and signs, and it takes a detailed JSON description of the scene as its prompt, which a helper step can write for you from a short sentence. The free weights are for non-commercial use only.

Summary

Ideogram, the company behind the Ideogram image service, published Ideogram 4 on 3 June 2026 as its first open-weight model. It has 9.3 billion parameters and, Ideogram says, was trained from scratch rather than fine-tuned from another model. It is a flow-matching model built on a single-stream diffusion transformer with 34 layers, in which text and image tokens share one sequence. Instead of CLIP or T5, it uses Qwen3-VL-8B-Instruct, a vision-language model, as its text encoder and reads the hidden states of 13 intermediate layers. It can draw any size from 256 to 2,048 pixels per side in steps of 16, with aspect ratios up to 6 to 1.

The model was trained only on structured JSON captions that spell out the scene, style, lighting, color palette, any text and, optionally, bounding boxes for each element. You can write that JSON yourself, and the README says it gives the most control. Plain prompts work through a magic-prompt step in the reference script, which by default calls Ideogram's hosted API with a free key and can instead use your own language model. The sampler presets run 48, 20 or 12 steps. The weights come as fp8, which runs on any hardware, and nf4, which needs an NVIDIA GPU with CUDA, and the table in the README lists Diffusers support for nf4 but not for fp8. The fp8 repository is about 27.5 GB and the nf4 one about 16.1 GB, each with a text encoder, an autoencoder and two transformer folders. The card gives no minimum VRAM. The script can also screen prompts and images with Hive's moderation services if you bring your own keys.

The weights are gated on Hugging Face and use the Ideogram Non-Commercial Model Agreement, which allows non-commercial use only. A company can test and prototype, but generating images for revenue-generating products is excluded, and a hosted service counts as distribution under the same terms. The inference code is Apache 2.0. Ideogram's own comparisons say Ideogram 4 is the best open-weight image model by a wide margin, ranks second overall behind GPT Image 2 medium on its internal designer-judged test, and renders text better than Qwen-Image, FLUX.2 [dev] and HunyuanImage 3.0. It also cites Design Arena, LMArena and a ContraLabs typography test, whose charts we could not check. All of these results are the developer's. In the 30 days to 11 November 2026, Hugging Face counted about 68,000 downloads of the fp8 repository.

Good for

  • Posters, logos, signs and other designs where the image must contain exact, legible text, including several lines
  • Layout control, with bounding boxes to place subjects and text and hex colors to steer the palette
  • Native 2K images and unusual shapes such as wide banners and phone wallpapers
  • Research and non-commercial projects that want a strong open text-rendering model

Watch out for

  • The license is non-commercial only, and generating images for revenue-generating products is excluded, so work use is limited to non-production testing
  • The weights are gated, and the best results need JSON prompts, so the default magic-prompt step either calls Ideogram's hosted API with a key or needs a language model of your own
  • No minimum VRAM figure is published, and the fp8 repository is about 27.5 GB, with the 16.1 GB nf4 build limited to NVIDIA CUDA
  • The fp8 build is listed without Diffusers support, so it runs through Ideogram's own code
  • The ranking and text-rendering claims are Ideogram's own comparisons

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Memory figures for each variant of Ideogram 4. Image, video, speech and music pipelines also load a text encoder and a decoder, so the memory needed can be larger than the main checkpoint.

Variants Parameters Download size Memory needed Fits on
Ideogram 4 (fp8) FP8 9.3B 28 GB 32 GB Gaming GPU
Ideogram 4 (nf4, NVIDIA CUDA) 9.3B 16 GB 19 GB Gaming GPU

Download size covers every component needed to run the model, such as the generator, text encoder and decoder. Memory needed is a stated or published measured minimum where one exists; otherwise it is the download size plus about 15%. Larger images, longer clips and higher resolutions need more.

The license in plain words

Ideogram's license for the Ideogram 4 weights, last updated on 3 June 2026. The weights are free only for non-commercial purposes, which the text defines narrowly. A company may test and prototype but not deploy, and generating images for products that earn revenue is excluded. Commercial use needs a separate agreement that Ideogram may grant or refuse as it likes.

  • Attribution or model name must be shown
  • Acceptable-use policy applies
  • Outputs must be filtered or reviewed, and AI generation disclosed

More about Ideogram Non-Commercial

See also

Frequently asked questions

What is Ideogram 4?
Ideogram 4 is an image model from the company behind the Ideogram image service. It is built to put exact, readable text into posters, logos and signs, and it takes a detailed JSON description of the scene as its prompt, which a helper step can write for you from a short sentence. The free weights are for non-commercial use only.
Can I use Ideogram 4 commercially?
Not without separate permission. Ideogram 4 is released under the Ideogram Non-Commercial, which allows research and personal use only.
Is Ideogram 4 open source?
Ideogram 4 is open-weight, not open source in the strict sense. You can download the model files, but the Ideogram Non-Commercial adds conditions that the Open Source Initiative definition does not allow.
Can I run Ideogram 4 on my own computer?
Yes, on a good graphics card. The smallest variant, Ideogram 4 (nf4, NVIDIA CUDA), needs about 19 GB of memory at 4-bit precision, which suits a 24 to 32 GB gaming GPU or a Mac with 32 GB or more.
When was Ideogram 4 released?
Ideogram 4 was released by Ideogram on June 3, 2026.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 11, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
Krea 2 Krea 12.8B – Krea 2 Community (Conditional) 41 GB Workstation
Qwen-Image-2.1 Alibaba Qwen 7.1B – Qwen Research License (Non-commercial) 20 GB Gaming GPU
FLUX 3 Image Klein Black Forest Labs 5.6B – FLUX Kommunity (Conditional) 8 GB Laptop or phone
Z-Image Tongyi MAI 6.15B – Apache 2.0 (Permissive) 16 GB Laptop or phone