NVIDIA has posted the Qwen-Image-NVFP4 model on Hugging Face. It is a quantized version of the Qwen‑Image text‑to‑image diffusion transformer, keeping the same architecture while lowering precision for faster GPU inference. The model supports up to 1,024 prompt tokens and runs on NVIDIA Blackwell GPUs with TensorRT‑LLM VisualGen. It is released under the Apache 2.0 license for commercial and non‑commercial use.
Why it matters
Developers can now generate images faster on NVIDIA GPUs without needing to quantize the model themselves.
