Photorealistic results
Alibaba highlights Z-Image’s photorealism: textures, light and fine detail close to a real photo. A good starting point for food shots, portraits or travel scenes.
Image model · Alibaba
Z-Image — Realistic images from a simple description.
Z-Image is an open-source, 6-billion-parameter image model from Alibaba’s Tongyi Lab, first released in November 2025. On Deepnia it turns a prompt of up to 1,000 characters into an image in one of 5 aspect ratios. Alibaba highlights its photorealism and accurate text rendering in English and Chinese.
- Text to image
- 1:1 · 4:3 · 3:4 · 16:9 · 9:16
- 1,000-character prompt
- Automatic resolution
- Photorealistic results
The image tool opens with Z-Image already selected. The exact cost shows before every creation; you sign up when you generate.
1:1
4:3
3:4
16:9
9:16
At a glance
- Maker
- Alibaba (Tongyi Lab)
- Input
- Text (your prompt)
- Aspect ratios
- 1:1, 4:3, 3:4, 16:9, 9:16 (1:1 by default)
- Resolution
- Automatic
- Prompt
- Up to 1,000 characters
- License
- Open source (Apache 2.0)
What Z-Image does best
Clear text in English and Chinese
According to Alibaba, Z-Image renders English and Chinese text accurately in the image, even in small type, for example on a poster or a shop sign.
Broad world knowledge
Alibaba says the model knows many famous places and everyday objects, which helps you get a believable scene from a short prompt.
An open model
Z-Image has 6 billion parameters. Alibaba publishes its code and weights under the Apache 2.0 license and presents it as an efficient foundation model.
One prompt, one ratio, done
On Deepnia, everything starts from your description: describe the image, pick the aspect ratio, then generate. Z-Image does the rest.
How to use Z-Image on Deepnia
Open the image tool
The “Try Z-Image” button opens Deepnia’s image generator with the model already selected.
Write your prompt
Describe the subject, setting, light and style in up to 1,000 characters: your description is all Z-Image needs to create the image.
Pick the aspect ratio
1:1, 4:3, 3:4, 16:9, 9:16; 1:1 is selected by default.
Generate
Check the cost shown, then click “Generate”. “Regenerate” runs the same request again for another version.
Z-Image on Deepnia: what you can set
- Input
- Text: your description is all it takes
- Prompt
- Required, up to 1,000 characters
- Aspect ratios
- 1:1, 4:3, 3:4, 16:9, 9:16 (1:1 by default)
- Resolution
- Automatic, in every aspect ratio
- Images per generation
- 1
- Text in the image
- Accurate in English and Chinese, according to Alibaba
Prompt ideas to try
Copy a prompt, open the tool and adapt it to your project.
Food photo for a restaurant menu
- Text only
- 1:1
Realistic food photo of a steaming bowl of ramen on a dark wooden table: golden broth, a halved soft-boiled egg, fresh scallions, chopsticks resting on the rim. Warm side light, blurred background of a small restaurant, high-end advertising style.
Packshot for an online store
- Text only
- 4:3
An amber glass perfume bottle on pale sand, surrounded by dried leaves, soft studio light, crisp reflections on the glass, terracotta gradient background. Advertising packshot style, perfectly sharp, no label or text.
Travel photo for a blog
- Text only
- 16:9
Realistic travel photo of a floating market at dawn: boats piled with tropical fruit, vendors in straw hats, light mist over the water, soft golden light. Wide angle, natural colors.
Editorial portrait for a fashion magazine
- Text only
- 3:4
Full-length portrait of a young woman in a loose sand-colored suit, posing on a cobbled street in evening light, shallow depth of field, fashion magazine editorial style, subtle film grain.
Photo for a science lesson
- Text only
- 16:9
Realistic documentary photo of a bright classroom: three students look at leaves under a microscope while their teacher shows them a botanical chart, potted plants on the windowsill, natural morning light. Soft colors, no readable text.
Café storefront with an English sign
- Text only
- 9:16
Front of a small neighborhood café at sunrise, forest-green facade, a large sign painted in white letters: "MORNING BREW", tables outside, steam rising from a cup on the counter. Realistic photo, soft light.
Tips for better results
- Write a concrete, ordered prompt: subject, setting, light, framing, then style, within 1,000 characters.
- For text in the image, write short words in quotation marks: Alibaba highlights accurate rendering in English and Chinese.
- For a long headline, ask for empty space in the image and add the text afterwards.
- Describe what you want to see rather than what you want to avoid.
- For a different composition, change one element of the prompt (angle, light, setting) rather than running the exact same request again.
- Pick the aspect ratio for the use: 1:1 for a post, 3:4 for a portrait, 9:16 for a story, 16:9 for a banner or an article.
Good to know
- To keep the same style across several images, reuse the same description of the setting, light and style, and change only the subject.
- For a believable scene from a short prompt, name a famous place or everyday objects: Alibaba says the model knows many of them.
- Each generation gives 1 image: click “Regenerate” for another version, or change one detail in the prompt to vary the composition.
- Proofread every word written in the image before you publish, accents included.
Z-Image or another model?
Deepnia’s other image models, to pick the right one for your project.
| Model | Resolution | Aspect ratios | References | Best for |
|---|---|---|---|---|
| Z-ImageThis model | Automatic | 5 ratios | None | Realistic images from a simple description. |
| GPT Image 2 | 1K–4K | 8 ratios | Up to 16 | Text-rich images: infographics, posters and graphic design. |
| Nano Banana 2 Lite | 1K | 14 ratios | Up to 10 | Deepnia’s default image model, to create or edit with several references. |
| Grok Image | Automatic | 5 ratios | Up to 1 | Editing a photo by simply describing the change. |
| Seedream v5.0 Lite | Automatic | 7 ratios | Up to 10 | 2K images, created or edited from several references. |
| Seedream v5.0 Lite Sequential | Automatic | 7 ratios | Up to 10 | A consistent series of 1 to 15 images in a single request. |
Frequently asked questions
Who made Z-Image?
The Z-Image team at Tongyi Lab, part of Alibaba. The model has 6 billion parameters and was first released in November 2025.
Is Z-Image open source?
Yes: Alibaba publishes the code and weights under the Apache 2.0 license. On Deepnia you use it right in your browser, with no installation or graphics card.
Z-Image or Grok Image: which one should I use?
Z-Image to create a photorealistic image from a simple description: a dish, a portrait, a travel scene. Grok Image to edit a photo you already have by describing the change in one sentence.
Can Z-Image write text in the image?
Yes: Alibaba highlights accurate text rendering in English and Chinese, even in small type, for example on a shop sign or a poster. Put the text in quotation marks, keep it short and proofread the result.
How do I get a truly realistic photo?
Describe the scene like a photographer: subject, setting, light (morning light, side light), framing (wide angle, close-up) and style (advertising photo, documentary). Alibaba highlights Z-Image’s photorealism: textures, light and fine detail.
Which aspect ratios are available?
1:1, 4:3, 3:4, 16:9, 9:16, with 1:1 by default. The resolution is set automatically.
How long can the prompt be?
Up to 1,000 characters on Deepnia, enough to describe the subject, setting, light and style.
How many images do I get per generation?
1 image per generation. For a consistent series of several images, use Seedream v5.0 Lite Sequential.
Ready to try Z-Image?
The image tool opens with Z-Image already selected. The exact cost shows before every creation; you sign up when you generate.