FLUX 3 Image Klein
Generate and edit images with downloadable weights, multiple references, bounding-box controls, and hosted API access.
Qwen-Image-2.1-Turbo
huggingface.co
An image generation and editing model with an eight-step sampling schedule, downloadable weights and hosted API access.
Visit website
Pricing
Paid
Available on
Diffusers, Downloadable model, Alibaba Cloud Model Studio API
Category
Ai Image Tools
What is Qwen-Image-2.1-Turbo?
Qwen-Image-2.1-Turbo is an accelerated image generation and editing checkpoint. It turns text instructions and image references into image outputs using an eight-step denoising schedule. Its visual generation component has 7 billion parameters.
The model serves researchers and developers building visual applications. Qwen provides downloadable weights and generation and editing examples, while Alibaba Cloud Model Studio offers hosted API access. The downloaded materials carry a research license, making licensing an important part of choosing between deployment routes.
Key features
Text-to-image generation
Generate images from written descriptions using the Turbo checkpoint's eight denoising steps.
Image editing
Use supplied imagery with editing instructions. The official repository includes Turbo-specific editing examples.
Diffusers integration
Load the checkpoint through QwenImage21Pipeline. The saved configuration includes the recommended sampling schedule and defaults to CFG=1.
Resolution presets
Documented formats include square, portrait and landscape presets, with examples such as 2048 × 2048 and 2752 × 1536.
Hosted image API
Alibaba Cloud Model Studio accepts text and image inputs and returns images, offering an alternative to managing checkpoint inference.
Who is it for?
Visual concept exploration
Design researchers can generate candidate visual directions from short text descriptions.
Reference image transformation
Visual workflow builders can explore requested changes to a supplied image.
Image application evaluation
Developers can compare the downloadable pipeline and hosted API for a proposed image feature.
How to use Qwen-Image-2.1-Turbo
- Review the research license and choose downloaded inference or the hosted Model Studio API.
- For downloaded inference, prepare CUDA-compatible PyTorch, current Diffusers source, transformers 5.17.0 or later, accelerate and Pillow.
- Load the Turbo checkpoint with QwenImage21Pipeline and retain its recommended sampling configuration.
- Follow an official generation or editing example with your prompt, optional reference image and a documented resolution preset.
Qwen-Image-2.1-Turbo pricing
Paid
Hosted API inference uses pay-as-you-go billing. Downloaded weights are available under the non-commercial Qwen Research License; commercial checkpoint use requires a separate license.
Model Studio API — Singapore International
USD 0.016 per generated image
Pay-as-you-go per image; no monthly interval
Standard Singapore International image-generation rate, excluding limited-time promotions. This is hosted inference pricing, separate from checkpoint licensing.
What to know before you use it
- The downloaded materials are licensed royalty-free for non-commercial research or evaluation only. Commercial use requires a separate commercial license; redistribution also carries attribution and license requirements.
- The recommended schedule is stored in the checkpoint. Changing num_inference_steps alone does not replace it, and alternative schedules have not been evaluated.
- The hosted service does not support function calling, web search, structured outputs, context caching or batch inference. Its documented rate limit is 120 requests per minute.
Frequently asked questions
Can I use the downloaded checkpoint commercially?
Commercial use requires a separate commercial license. The Qwen Research License covers non-commercial research and evaluation.
Do I need to host the model myself?
No. The official project also links to hosted Turbo API access through Alibaba Cloud Model Studio.
Does local prefix caching imply hosted context caching?
No. The local pipeline supports prefix KV caching, while Model Studio lists hosted context caching as unsupported.