Ideogram 4.0, an image generation AI that surpasses Nano Banana Pro and can run locally, has been released and can also be controlled in Japanese.
AI company Ideogram released its image generation AI, ' Ideogram 4.0, ' as an open model on June 3, 2026. Despite being an open model that can run locally, Ideogram 4.0 has been evaluated by third-party organizations as having performance exceeding Google's Nano Banana Pro.
Ideogram 4.0 | Ideogram
Ideogram 4.0 Technical Details: Open model at the forefront of design
https://ideogram.ai/blog/ideogram-4.0/
Ideogram 4.0 is an image generation model with 9.3 billion parameters that can generate both realistic and illustrative images. Below is an image showing examples of images generated by Ideogram 4.0.
It excels particularly in text rendering, and can accurately render long sentences.
Furthermore, by dividing the objects and text within the training images into bounding boxes, we successfully enabled the model to understand the 'positional relationships between objects and text.' This makes it easier to provide layout instructions via prompts.
The following is a ranking of image generation AIs by the third-party organization DesignArena. Ideogram 4.0 ranked 4th in the world, surpassing Nano Banana Pro. DesignArena created the ranking using a method in which 'humans evaluate the quality of the generated output without knowing the name of the AI,' meaning that Ideogram 4.0 was evaluated as high-performance by human eyes rather than by mechanical benchmarks.
In a ranking focused solely on open models, Ideogram 4.0 secured the top spot, significantly ahead of second-place HunyuanImage-3.0 and third-place FLUX.2 [dev].
The graph below shows the number of parameters, which represents the model's size, on the horizontal axis, and character rendering performance on the vertical axis. Ideogram 4.0 boasts the best character rendering performance among open models while keeping the number of parameters relatively small.
A free demo app is available, so I tried it out.
Ideogram 4 - a Hugging Face Space by ideogram-ai
Generated by entering 'A maid sitting on an air conditioner unit in an alley, reading a newspaper. Japanese woman. Black hair in pigtails.'
The generated image is shown below. It can handle instructions in Japanese perfectly. This time, it turned out in an illustration style even though no specific instructions were given.
Ideogram 4.0 is available in FP8 and NF4 versions at the following link. The NF4 version is sized to fit on a GPU with 24GB of VRAM.
ideogram-ai/ideogram-4-fp8 · Hugging Face
https://huggingface.co/ideogram-ai/ideogram-4-fp8
ideogram-ai/ideogram-4-nf4 · Hugging Face
https://huggingface.co/ideogram-ai/ideogram-4-nf4
Furthermore, image generation using Ideogram 4.0 is already possible with ComfyUI.
Ideogram 4.0 is now natively supported on ComfyUI @ideogram_ai v4.0 is an open-weight 9.3B text-to-image foundation model.
— ComfyUI (@ComfyUI) June 3, 2026
It is exclusively trained on structured JSON caption datasets for precise scene description.
It features flawless text rendering, professional layout… pic.twitter.com/132RCXxuD0
The source code and related documentation are also available at the following link, and Ideogram is looking forward to community contributions to improve the model. They also plan to release an image editing model in the near future.
GitHub - ideogram-oss/ideogram4: Ideogram 4: Open image model at the forefront of design · GitHub
https://github.com/ideogram-oss/ideogram4
Related Posts:
in AI, Posted by log1o_hf