Qwen Image 2.1 AI Image Generator

What Qwen Image 2.1 does well

Transparent PNG images, made directly

Qwen Image 2.1 can return an image with a real transparent background. A sticker, an icon or a product cut-out is ready to drop onto a slide, a web page or another photo, with no separate background-removal step.

Set the Background option to transparent and describe only the subject: a cartoon fox sticker, a glass perfume bottle, a hand-lettered word.

Qwen also shows the model editing images that are already transparent, such as changing a character's expression while keeping the empty background, and pulling one subject out of an everyday photo onto its own layer.

This is an RGBA image with transparency. Three die-cut cartoon stickers in a row: a steaming coffee cup, a golden croissant and a smiling sun, each with a thick white border, flat bright colors and a soft drop shadow. The image has alpha channel and the background is transparent.

Combine multiple images into one picture

The model reads several reference images in one request and keeps track of which is which.

Qwen's own examples:

• Six separate portraits put into one group photo.
• A model dressed in a coat, shoes, a bag and a hat taken from five different product shots.
• A whole room furnished from ten pictures of furniture.

On Zero Kit you can upload up to 10 images. Refer to them by their order, as in "the person from image 1 wearing the jacket from image 2", and say what each image is for.

To change the clothes in one photo of a person, the AI outfit swap tool does that job on its own.

A side-by-side diptych with a thin white divider. Left half: a neat grid of five reference photos on white backgrounds, a camel wool coat, a pair of white leather sneakers, a small brown leather bag, a cream knit beanie, and a young model standing in a plain grey t-shirt and jeans. Right half: the same model wearing the coat, sneakers, bag and beanie, walking on a city street in autumn, every item matching its reference. Natural light, realistic photography, no text, no logos.

Local edits that keep faces and products

For a precise edit, mark the area on the photo before you upload it: draw a colored circle around the object or paint over the region, then say what the mark means.

Qwen's examples use three colored circles in one photo to remove a watch, change hair color and swap a shirt in a single edit.

Qwen says this version holds on to a person's facial features and to a product's printed text, texture and shape better from input to result, which is what portraits and product shots need.

A side-by-side diptych with a thin white divider, showing the same bright living room from the same angle. Left: a grey linen sofa, a wooden coffee table, a brass floor lamp beside the sofa with a hand-drawn red circle around the lamp. Right: the identical room with the lamp replaced by a tall fiddle-leaf fig in a white ceramic pot, the red circle gone, everything else unchanged. Soft daylight, realistic interior photography, no text, no logos. Identical framing in both halves: the same single grey sofa, coffee table, rug and windows; only the lamp is replaced.

How to write prompts for Qwen Image 2.1

1. For an image with a transparent background, switch the Background option to transparent, describe only the subject, and leave out words like "on a table" or "in a room".

2. With several references, name each image by its order and its role, for example "image 1 is the model, image 2 is the coat, image 3 is the shoes".

3. For a local edit, use a different color for each marked area and give one instruction per color, such as "remove what is in the blue circle".

4. Put every word that must appear in the image in quotation marks, keep it short, and check the spelling at full size before you use the result.

5. Short prompts are expanded into a fuller description before generation, so a brief idea works; add detail only where you want control, such as the light, the material or the layout.

Sample prompts to try with Qwen Image 2.1

• 

  A sign with clean lettering

  A neon shop sign that reads "NIGHT BAKERY" in pink cursive letters with a blue outline, mounted above a small bakery window on a rainy night street, loaves visible behind the glass, glowing reflections on the wet pavement, light drizzle, square composition, correct spelling, no other text, no logos.

• 

  Portrait in window light

  Vertical portrait of a violin maker in her workshop, a woman in her forties with tied-back grey hair and a canvas apron, sanding the spruce top of an unfinished violin. Soft side light from a tall window, visible skin texture and fine wood grain, chisels and clamps on the bench, curls of wood shavings, warm natural tones, 85mm lens, no text, no logos.

• 

  Infographic in four steps

  A clean horizontal infographic on a warm cream background titled "FROM BEAN TO CUP" at the top in bold letters. Four illustrated steps from left to right, joined by thin arrows: a red coffee cherry on a branch labeled "HARVEST", green beans drying on a rack labeled "DRY", brown beans in a roasting drum labeled "ROAST", a cup of coffee labeled "BREW". Flat illustration style with soft shadows, correct spelling, no other text, no logos.

• 

  Full outfit in one shot

  Full-length street-style fashion photo of a young person walking on a city sidewalk in autumn, wearing a camel wool coat, a cream knit beanie, straight blue jeans, white leather sneakers and a small brown leather shoulder bag. Red brick buildings and fallen yellow leaves behind, soft overcast light, natural candid pose, vertical frame, no text, no logos.

Sample images are illustrative; results vary by model.

Qwen Image 2.1 specs and pricing

Every price below comes from the live price list, so it changes here as soon as rates change. Credits are charged per image, or per second of video.

Qwen Image 2.1 limitations

• A transparent background works best when the prompt describes only the subject. If the prompt mentions a scene, a table, a wall or any other surface, the result may come back with a background even when transparency is switched on.

• Every extra reference image gets a smaller share of the model's attention, so fine details such as small logos or printed patterns can drift when you upload many at once.

  For a faithful product or face, use fewer references and say exactly what each one is for.

• The editor has no mask brush for this model. To change one area, draw a circle or paint over it on the photo before you upload it, and describe in the prompt what the mark asks for.

  The marks cover part of the original picture, so keep them thin and only around what should change.