Qwen Image 2.1 Turbo

Turbo accelerates the Qwen Image 2.1 family's 7B visual generation component. Its official Diffusers example uses the saved eight-step schedule with CFG=1; changing num_inference_steps alone does not override that schedule. Start with the checkpoint's documented configuration before comparing timing or quality.

The hosted API supports text-to-image and editing, up to ten reference images, one to six outputs per call and resolutions through 2048×2048. Qwen documents transparent RGBA output and local edits. These are API/model contracts; a reference limit does not establish perfect identity or geometry preservation.

Do not relabel this checkpoint as Qwen Image 3.0. The hosted IDs and documented reference limits differ: 2.1 supports up to ten references, while the 3.0 guide lists up to three. Pro and Turbo are also distinct IDs; check the selected variant rather than importing another page's defaults.

The published Qwen Research License defines non-commercial use as research or evaluation and requires a separate commercial license for commercial model use. It covers model materials and documentation; public download availability is not an Apache 2.0 grant. Hosted Alibaba API terms and regional billing must be checked separately. The reviewed international listing advertises $0.016 per image and 120 requests per minute; those are that surface's rates, not a universal entitlement.

The repository shows poster creation, a yacht-sketch edit, multi-reference edits and transparency. Those publisher examples illustrate intended jobs; they are not independent evidence of quality, memory use or production reliability. Community questions focus on quantization, sampler/CFG choices and ComfyUI compatibility. Compare the same inputs and settings on your own licensed surface before adopting a workflow.