Ideogram 4
#33 in Text-to-BildIdeogram · v4 · siet 3. Juni 2026 · 6× · tolest 29. Juni 2026
6
Momentum
Ideogram 4.0 is a text-to-image model developed by Ideogram AI, released on June 3, 2026 as the company's first open-weight model. It is a 9.3-billion-parameter single-stream Diffusion Transformer with 34 layers, trained from scratch and paired with a frozen Qwen3-VL-8B vision-language model as text encoder. The model stands out for highly accurate text rendering inside images, structured JSON prompting with bounding-box layout control, and native resolutions up to 2048 pixels. Weights are available via Hugging Face and GitHub; the inference code is Apache 2.0 licensed, while the model weights themselves fall under a non-commercial license.
Momentum-Verloop
16.05.14.08.
Features
| Image Resolution (Max.) | Native 256–2048 px per side (multiple of 16), maximum aspect ratio 6:1; default: 2048×2048 px (native 2K) |
| Fine-Tuning | Fine-tuning/LoRAs permitted for non-commercial use; commercial fine-tuning and custom model training via Enterprise/Self-Serve |
| License | Inference code: Apache 2.0; model weights: Ideogram Non-Commercial Model Agreement (commercial use requires a separate paid license) |
| Max Resolution | Native resolution from 256 to 2048 pixels per side, flexible aspect ratios |
| Platform | Web app (ideogram.ai), API, MCP, Hugging Face, GitHub, self-hosting on own hardware |
| Price | Subscription from $7/month (Basic); API from $0.03 to $0.10 per image depending on quality tier |
| Release Date | June 3, 2026 |
| Memory Footprint (GB) | ~20–22 GB VRAM (FP16 backbone); nf4 checkpoint (quantized) runs on a single 24 GB GPU (e.g., RTX 4090); fp8 checkpoint recommends 40–80 GB GPU for production batch inference |