Ideogram 4.0 Open Image Model Tops Open Rankings, Runs Locally
The release of a high-performing open image model that runs locally and beats a major proprietary competitor in human evaluation shifts the balance toward accessible, community-developable AI tools.
Reporting from 1 sources: GIGAZINE.
Ideogram released Ideogram 4.0 as an open model on June 3, 2026. The 9.3-billion-parameter model ranks 4th overall and 1st among open models on the DesignArena leaderboard, surpassing Google's Nano Banana Pro. It excels at text rendering and layout control, supports Japanese instructions, and can run on a 24GB VRAM GPU. A free demo and ComfyUI integration are available.
Ideogram 4.0, an open image generation model with 9.3 billion parameters, was released on June 3. On the DesignArena leaderboard, which ranks models by human evaluation, it placed 4th overall and 1st among open models, beating Google's Nano Banana Pro and open competitors HunyuanImage-3.0 and FLUX.2 [dev] by a wide margin. The model is designed for strong text rendering and layout understanding, using bounding boxes to separate objects and text during training. It supports Japanese-language prompts and can generate both photorealistic and illustration-style images. Two quantized versions are available: FP8 and NF4, the latter fitting into a GPU with 24GB VRAM. Ideogram has also published source code and documentation on GitHub, inviting community contributions, and plans to release an image editing model soon. A free demo is running on Hugging Face Spaces, and ComfyUI already supports generation with the model.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.