Qwen-Image
#17 v Text na obrázekalibaba · od 2025-08-05 · 5× · naposledy 29. 6. 2026
22
Momentum
Qwen-Image is a 20-billion-parameter image generation model (MMDiT architecture) by Alibaba's Tongyi Qianwen team, open-sourced in August 2025 under Apache 2.0. It was specifically designed for high-fidelity multilingual text rendering (especially Chinese and English) and achieved first place across 9 public benchmarks at launch. The model supports both text-to-image generation and precise image editing. It was succeeded by Qwen-Image-2.0 (February 2026), which uses 7B parameters and native 2K resolution.
Vývoj momenta
04.04.03.07.
Vlastnosti
| Memory Footprint (GB) | ~61.8 GB VRAM at FP16 precision (1024×1024); model file size ~57 GB (BF16); deployable on a single RTX 3090 with DFloat11 quantization + CPU offloading; FP8 download ~26.7 GB |