Soutine Qwen Image Prompts
Prompt library: https://soutine.ai/qwen-image-2-1-prompts
Soutine's Qwen Image 2.1 prompt library is a focused set of examples for creating images and working with references. It covers portrait and product scenes, transparent assets, image edits, multi-reference compositions, and creative layouts.
The live landing page helps you choose an example, inspect its full prompt and input requirements, then prepare the Qwen Image generator. This Hub repository is a reference card, not a distribution of model weights or a local image-editing package.
Library contents
At the time of this update, the page presented 12 examples. Its categories illustrate different tasks rather than a single repeated prompt format:
| Category | Examples shown on the page | Input considerations |
|---|---|---|
| Portraits | Editorial and vintage-film portraits | Some examples begin from text without a reference. |
| Products | Studio product photography | Describe the product, lighting, and composition. |
| Transparent assets | Cat sticker and isolated handbag | Use the generator's transparency setting and inspect the original PNG. |
| Image editing | Outfit changes and background replacement | Supply the subject reference required by the template. |
| Multiple references | An outfit from five references; three portraits in one scene | Give each input a clear role and keep its numbering consistent. |
| Creative design | Coffee poster, infographic, and miniature bakery | Specify layout and quote any exact text you want. |
Cards expose Full prompt, Copy prompt, and Use this prompt, along with reference-image requirements. The collection and generator controls may change over time.
Intended use
The library can help designers preparing a product composition, creators exploring portrait treatments, and teams working on draft posters or infographic layouts. Its editing and multiple-reference examples are particularly useful for learning how to assign separate roles to a person, garment, accessory, or background.
Transparent-asset examples provide a starting point for stickers and isolated product visuals. They still require inspection for unwanted background pixels, edge quality, and actual alpha transparency before use in a design.
How to use a template
- Choose the task category. Decide whether you need a text-generated image, an edit, a transparent asset, or a composition using several inputs.
- Read Full prompt and the reference requirements. Check whether the example expects no reference, one subject image, or several inputs with specific roles.
- Copy or choose Use this prompt. The latter fills the generator; it does not start generation automatically.
- Replace the example details. Describe your desired finished image. For editing, state what should change and what you want to preserve.
- Upload and map references. Keep their order aligned with the prompt's numbering. Editing templates require their reference images before submission.
- Check output settings and quote. Select the available resolution and, for a transparent asset, the Transparent setting. Confirm the current credit cost before submitting.
- Inspect the original output. Check identities, garments, object shapes, spelling, layout, and transparency as applicable. Download the original file when alpha transparency matters.
Reference roles and exact wording
The page's multiple-reference templates distinguish roles such as person, jacket, trousers, shoes, and handbag. Use clear assignments rather than expecting the model to infer which detail to take from each image. Upload order matters because it determines reference numbering.
For posters, place desired words in quotation marks and state their intended position. This improves the clarity of the instruction but does not guarantee correct lettering or exact layout.
Illustrative reference-edit prompt
This is an original teaching example, not a tested prompt or a claim about a generated result:
Use reference 1 for the person and pose, and reference 2 for the jacket. Show the person wearing the jacket in a simple studio portrait. Keep the camera framing and background close to reference 1. Do not add accessories or lettering.
The two inputs have different roles. After generation, compare the face, pose, jacket details, and background with the references rather than assuming those instructions were followed exactly.
Output and input considerations
The current landing-page FAQ describes 1K and 2K outputs, one image per generation, and no 4K option for this workflow. It also describes PNG, JPEG, and WebP references with size and image-count limits. Check the live generator for the current limits before preparing an upload.
For transparent output, select Transparent and describe the desired RGBA image. The FAQ states that outputs are PNG and that the original download preserves the alpha channel. Some preview illustrations may be JPG, so a preview alone is not evidence of transparency in an output file.
Limitations
Prompts express desired behavior, not guaranteed fidelity. Identity, clothing details, object boundaries, perspective, text, and reference consistency can vary. A longer or conflicting instruction can also change parts of the image you intended to preserve.
Multiple references require careful ordering and inspection; assigning a role does not ensure exact reproduction of every input detail. Transparent assets may need edge cleanup, and generated posters or infographics need a character-by-character text review.
Generation takes place on Soutine and uses its credits and current terms. Use reference material you have permission to upload, and check commercial-use eligibility and output visibility before publishing.
Frequently asked questions
Do all examples require an upload?
No. The page marks examples with no required reference and others with specific reference roles. Check the selected card.
Does Use this prompt immediately spend credits?
No. It fills the generator. You still provide required inputs, review the quote, and submit the task.
How should I request transparency?
Choose the Transparent setting, describe the intended asset, and inspect the original PNG's alpha channel after generation.
Can I use these as prompts for another model?
They can be a starting point, but controls, reference handling, and results differ. The examples do not establish equivalent behavior elsewhere.
Does the Hugging Face card include a runnable Qwen model?
No. It contains a README pointing to Soutine's prompt library and external workflow.
Explore the templates
Open Qwen Image 2.1 prompts to compare tasks and reference requirements before choosing an example.
Links
About this repository
The prompt library lives on the landing page. This is a README-only reference card for discovery and deep linking. It contains no model weights, training code, or hosted inference endpoint.