Alibaba Open-Sources Qwen-Image-2.1-Turbo: 2K Image Generation in 8 Denoising Steps
Alibaba's Tongyi Qwen team released and open-sourced Qwen-Image-2.1-Turbo on October 9. It is a sped-up version of Qwen-Image-2.1, with weights available on Hugging Face under Qwen/Qwen-Image-2.1-Turbo and the API launched alongside it. The model dramatically improves generation speed while retaining the original model's capabilities, marking another significant update for open-source image generation.
Confirmed
- The model is an accelerated checkpoint of Qwen-Image-2.1, using accelerated distillation to compress denoising from dozens of steps down to 8 steps, with no loss in quality per the team
- Retains the same 7B vision generation architecture as the original, supporting text-to-image and image editing
- Still generates 2K resolution images and supports image editing via natural language instructions
- Defaults to CFG=1 and adds KV caching for further speed gains
- Available through both diffusers and official usage paths, with weights open for download and the API live simultaneously
Why it matters
- 8-step denoising sharply cuts inference cost and wait times, bringing high-quality 2K image generation closer to practical, even real-time, use cases
- Fully open-source (downloadable weights) plus an API channel serve both local deployment and cloud calling needs
- In competition with closed-source image generation services, it gives the community an open-source option balancing speed and quality
2026-10-09 ~ 2026-10-09 · 9 related posts
Primary sources
- Alibaba open-sources Qwen-Image-2.1-Turbo, generating 2K images in just 8 denoising steps — Alibaba_Qwen ·
- Qwen-Image-2.1-Turbo Lands on Hugging Face With 8-Step Sampling, CFG=1 and Prefix KV Caching — SysPsych ·
- Qwen releases 8-step accelerated Qwen-Image-2.1-Turbo on Hugging Face — multimodalart ·
- Qwen releases Qwen-Image-2.1-Turbo text-to-image model on Hugging Face — Qwen · 2026-10-09
- [source] Alibaba open-sources Qwen-Image-2.1-Turbo, generating 2K images in just 8 denoising steps — Alibaba_Qwen · 2026-10-09
- [source] Qwen-Image-2.1-Turbo Lands on Hugging Face With 8-Step Sampling, CFG=1 and Prefix KV Caching — SysPsych · 2026-10-09
- Qwen Releases Qwen-Image-2.1-Turbo: Open-Weights Image Model That Needs Only 8 Denoising Steps — ResearchCrafty1804 · 2026-10-09
- Alibaba's Qwen releases Qwen-Image-2.1-Turbo on Hugging Face — chocofoxy · 2026-10-09
4 near-duplicate retellings: CeFurkan · CeFurkan · multimodalart · CurieuxExplorer