Today we published a technical blog post about Ideogram 4.0 — our goal is to enable more innovation and creativity. It's a 9.3B Diffusion Transformer trained from scratch, paired with a frozen 8B VLM as text encoder. The nf4 checkpoint runs on a 24GB consumer GPU. Thread 🧵 https://t.co/MeG7vrDzgt
Ideogram Releases 4.0 Open Weights for Local 2K Graphic Design
Ideogram· Updated
Ideogram has released the weights for Ideogram 4.0, a 9.3 billion parameter Diffusion Transformer model paired with an 8B VLM text encoder. This open-weight release enables local hosting and fine-tuning on private data, running on a 24GB consumer GPU, and ranks as the top open-weight model on third-party leaderboards, closing the gap with proprietary systems.
- Model Architecture
- 9.3B Diffusion Transformer
- Text Encoder
- Frozen 8B VLM
- GPU Requirement
- 24GB consumer GPU (nf4 checkpoint)
- Max Resolution
- Native 2K
- Leaderboard Rank
- #1 Open-Weight (DesignArena)
Ideogram 4.0 is the #1 open-weight text-to-image model in the world on DesignArena's third-party leaderboard. It is ahead of every foundation model except closed models from OpenAI and Google, offering a high-fidelity alternative for teams requiring data privacy or custom model behavior through local hosting.
The weights are downloadable for fine-tuning (adapting a pre-trained model to a specific dataset) or local inference, with the nf4 checkpoint fitting a 24GB consumer GPU. Ideogram 4.0 is live on the Ideogram web platform and API, plus partners including Hugging Face, ComfyUI, and fal, with the code on GitHub.
Still wondering? A few quick answers below.
Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →






