Alibaba open-sources Qwen-Image-2.1-Turbo, generating 2K images in just 8 denoising steps
Alibaba_Qwen · x · 2026-10-09
Alibaba's Qwen team released and open-sourced Qwen-Image-2.1-Turbo, an accelerated checkpoint of Qwen-Image-2.1 built on the same 7B visual generation architecture, cutting generation down to 8 denoising steps with no quality trade-off: it still produces strong 2K text-to-image output and supports natural-language follow-up edits (adding accessories, changing scenes).
Key points:
- Open weights are on ModelScope and Hugging Face; in Diffusers, load QwenImage21Pipeline and the recommended 8-step schedule works out of the box
- Qwen-Image-2.1 Pro and Turbo hosted APIs are now live, alongside self-hosted Turbo weights
- The quoted launch post on 2.1: a compact 7B model claimed to beat most closed-source rivals, with native RGBA layer generation/editing, up to 10 reference images, and high-fidelity local edits
Related event: Qwen Open-Sources Qwen-Image-2.1-Turbo: 2K Images in 8 Denoising Steps(6 posts)→
More from Multimodal
- This Workflow Uses Astra to Edit Video Autonomously: 20 Mins Setup, 2 Mins of Feedback per Hour — nickbaumann_ · 2026-10-09
- Kling 4.0 Flash earth-zoom effect goes viral, creator shares tweaked prompt — umesh_ai · 2026-10-09
- YarnGPT relaunches with AI speech and video dubbing for African languages — saheedniyi_02 · 2026-10-09
- Running MiniMax-H3 video gen on an 8GB VRAM laptop, now asking for text-to-image picks — Zoaloo · 2026-10-09
- Synthesia launches Syren Video: prompt-to-branded AI video with conversational editing — synthesiaIO · 2026-10-09
- Seedance 2.5 used to create a 30s AAA-style survival gameplay concept with prompt — azed_ai · 2026-10-09