Model
Qwen3 AI Image Generator
Qwen3 is an Alibaba image model offered here in standard and Pro versions. Its appeal is control: it accepts negative prompts and a fixed seed, and it reads prompts written in Chinese or English. Both versions support text-to-image and edits with up to three reference images.
What Qwen3 does well
Negative prompts
You can list what to exclude, up to 5,000 characters, which helps remove recurring unwanted elements such as stray text or clutter.
Seed control
Setting an integer seed makes it easier to compare small wording changes side by side instead of starting from a fresh random result each time.
Chinese and English prompts
Prompts can be written in either language, which helps bilingual teams and content aimed at Chinese-speaking audiences.
Your exact wording
OpenDream sends your prompt as written, with the model's automatic prompt rewriting switched off, so changes you make are the changes the model sees.
Best for
- Iterating on one composition with a fixed seed
- Excluding unwanted elements with negative prompts
- Writing prompts in Chinese or English
- Light edits guided by one to three reference images
- Widescreen 21:9 or vertical 9:16 images at 1K or 2K
Prompt examples
Seeded Typewriter Still Life
tabletop scene of a vintage typewriter with a half-written letter, dried flowers in a glass jar, afternoon sunlight through lace curtains, soft shadows, warm nostalgic tones
Fix the seed, then change one detail at a time, such as the flowers or the light, to see its effect in isolation.
Try this promptClean Bakery Storefront
small independent bakery storefront with a pale green facade, large window displaying bread loaves, striped awning, soft morning light, empty sidewalk, straight-on architectural view
Add a negative prompt such as 'people, cars, clutter, watermark' to keep the facade clear.
Try this promptInk-Wash Pavilion in Chinese
水墨风格的山间小亭,清晨薄雾,远处有瀑布,淡灰色调中点缀一抹青绿,留白构图
This prompt describes an ink-wash mountain pavilion at dawn with mist and a distant waterfall; writing it in Chinese is fully supported.
Try this prompt
Frequently asked questions
- Does Qwen3 support negative prompts and seeds?
- Yes. Both Qwen3 and Qwen3 Pro accept a negative prompt of up to 5,000 characters and an integer seed.
- Can I write Qwen3 prompts in Chinese?
- Yes. The model accepts prompts in both Chinese and English.
- How many reference images does Qwen3 accept?
- Image-to-image requests take one to three reference images. Upload PNG, JPEG, WebP or AVIF; OpenDream resizes large files for the model.
- What is the difference between Qwen3 and Qwen3 Pro?
- They share the same inputs and limits, including 1K or 2K output. Pro costs more credits per image, so compare both on your own prompt before settling on one.
Related models
Try Qwen3
Or let Smart mode choose the best model for each request.
Start creatingCapabilities in OpenDream
| Capability | Qwen3 Pro |
|---|---|
| Text to image | Yes |
| Edit / image to image | Yes |
| Reference images | Up to 3 |
| Resolutions | 1K · 2K |
| Aspect ratios | 8 |
| Negative prompt | Yes |
| Seed | Yes |
| Transparent background | No |
| Speed | Medium |
| Quality | High |
| Credits per image (from) | 7 |
Live from the OpenDream model registry. Costs can vary with resolution and reference images; the exact cost is shown before you generate.