Qwen Image 2.1 Turbo
An open-source accelerated image generation model released by the Qwen team, built on Qwen Image 2.1 with optimized inference speed. It generates images in as few as 8 denoising steps, substantially reducing the number of sampling steps required compared to its predecessor.
Timeline (last 90 days)
The official Turbo version is an 8-step accelerated version, and when using it you need to set steps to 8, CFG=1, Shift=6.5 and disable resolution dependency.
Stacking Turbo LoRA on the Turbo model can compress generation to 5-6 steps while maintaining overall quality.
Qwen Image 2.1 Turbo requires only 8 denoising steps to generate images, reducing sampling steps by 80% compared to the original version's approximately 40 steps.
Qwen Image 2.1 Turbo is based on a 7-billion-parameter architecture, can generate 2K images from text prompts and supports natural language image editing, with weights publicly available on Hugging Face.
All Facts (4)
The official Turbo version is an 8-step accelerated version, and when using it you need to set steps to 8, CFG=1, Shift=6.5 and disable resolution dependency.
50%UnverifiedStacking Turbo LoRA on the Turbo model can compress generation to 5-6 steps while maintaining overall quality.
50%UnverifiedQwen Image 2.1 Turbo requires only 8 denoising steps to generate images, reducing sampling steps by 80% compared to the original version's approximately 40 steps.
50%UnverifiedQwen Image 2.1 Turbo is based on a 7-billion-parameter architecture, can generate 2K images from text prompts and supports natural language image editing, with weights publicly available on Hugging Face.
50%