Designers and marketers
Explore poster, product-label and campaign concepts with a defined headline and composition before preparing the final production artwork.
MOSS-Image AI Image Generator
MOSS-Image generates photographs, illustrations and text-rich graphics from written descriptions. Its model family supports English and Chinese lettering and structured JSON layouts for arranging text and objects.
Text to image
Bilingual lettering
Layout control
Multiple styles
MOSS-Image generator — coming soon
Image generation on this site is not yet available. Explore the model, examples and supported settings below.
What Is MOSS-Image?
An open-weight image model family built for visual creation with text and spatial structure.
OpenMOSS introduced MOSS-Image on October 16, 2026. The family includes a pretrained Base checkpoint and an SFT checkpoint refined on curated data, sharing a 5-billion-parameter diffusion backbone.
MOSS-Image renders images from text. A prompt can describe the finished scene, spell out words to appear in the image, or specify a structured layout. The Base and SFT checkpoints are two versions of the same release, rather than successive generations.
The model family offers three complementary ways to describe an image:
- Describe the scene: Use natural-language descriptions or tags to specify subjects, lighting, materials and visual style.
- Write the visible text: Include the exact English or Chinese wording and explain its hierarchy, appearance and line breaks.
- Specify a layout: Use experimental JSON layout prompts to describe objects, text and their bounding boxes through the text input.
Layout generation creates a new raster image. It is not pixel-based editing of an uploaded photograph.
MOSS Image is an independent site and is not affiliated with OpenMOSS.
MOSS-Image Examples
Published MOSS-Image SFT examples from Sogni illustrate model capabilities. These are third-party demonstrations, not outputs generated on this site.
MOSS-Image
Handwritten English whiteboard
MOSS-Image
Concert poster
MOSS-Image
Bilingual tea menu
MOSS-Image
Product label
MOSS-Image
Three-panel comic
MOSS-Image
Portrait with lettering
What You Can Create with MOSS-Image
Explore six capabilities across lettering, composition and image style.
English and Chinese lettering
Combine readable text with visual scenes, from bilingual menus to illustrated titles. Check spelling, punctuation and line breaks before publishing.
Text-rich compositions
Describe headings, subheadings and smaller annotations together to build posters, educational graphics and editorial concepts.
Structured layout control
The model family accepts JSON prompts that assign positions and sizes to text and objects. This experimental control guides composition without guaranteeing exact placement.
Photographic scenes
Specify lighting, viewpoint, materials and background details for product concepts, portraits and other photographic-style images.
Illustration and comics
Create illustrated characters, anime-style scenes and multi-panel comic concepts with text and visual elements in one composition.
Flexible visual formats
Choose square, portrait or landscape compositions for different placements. Adjust the description and text hierarchy to suit each canvas.
Who Can Use MOSS-Image?
A starting point for creators who need images with clear visual and written information.
Educators and presenters
Develop illustrated explanations, whiteboard concepts and educational graphics. Review every generated label and factual statement.
Artists and storytellers
Sketch illustrated scenes, character moments and comic-panel arrangements with dialogue and visual direction in the prompt.
Publishers and independent creators
Prototype book covers, editorial graphics and social images that combine title typography with an original visual concept.
How to Generate Images with MOSS-Image
Build a clear prompt, choose a canvas and review the result.
1
Describe the finished image
State the subject, style, setting and composition. Include the exact words you want rendered and describe where they belong.
2
Choose the image settings
Select an aspect ratio and JPEG, PNG or WebP output. Use a square, portrait or landscape canvas that fits your composition.
3
Generate and review
Inspect text, object placement and visual details in the result. Refine the prompt as needed, then download the image.
MOSS-Image Specifications
Model identity and the image settings configured for MOSS Image.
| Developer | OpenMOSS |
| Model type | Text-to-image; raster output |
| Diffusion backbone | 5 billion parameters; text encoder and VAE are additional components |
| Image input | Text prompts; no uploaded reference images |
| Aspect ratios | 1:1, 4:3, 3:4, 16:9, 9:16 |
| Output dimensions | 1024 × 1024, 1024 × 768, 768 × 1024, 1024 × 576, 576 × 1024 |
| Download formats | JPEG, PNG, WebP |
| Prompt length | Up to 5,000 characters in this generator |
| Native photo editing | Not supported |
| Video generation | Not supported |
Last updated: October 22, 2026
MOSS-Image Base vs. SFT
The two official checkpoints share an architecture and input interface. These are model-family differences, not a checkpoint selector in this generator.
| Spec | MOSS-Image-Base | MOSS-Image-SFT |
|---|---|---|
| Role | Pretrained foundation | Fine-tuned from Base on curated data |
| Diffusion backbone | 5 billion parameters | 5 billion parameters |
| Text-to-image | Supported | Supported |
| JSON layout prompts | Supported; experimental | Supported; experimental |
| Direct reference-image input | Not supported | Not supported |
| Official whiteboard example | Less organized text grouping | Clearer text grouping and layout |
MOSS-Image-Base Advantages: Provides a pretrained starting point for customization and fine-tuning while retaining broad text-to-image capabilities.
MOSS-Image-SFT Advantages: Adds curated-data fine-tuning, with clearer organization in the official whiteboard comparison.
Consider Base for model adaptation and SFT for general generation. The published comparison illustrates one example and does not establish a winner for every task.
FAQs
MOSS-Image Questions and Answers
Understand image inputs, text rendering, layout prompts and output settings.
1
What is MOSS-Image?
MOSS-Image is an open-weight image model family from OpenMOSS. It generates photographs, illustrations and text-rich graphics from text prompts, with Base and SFT checkpoints.
2
Can MOSS-Image edit an uploaded photograph?
The released model backbone does not directly accept reference images or perform native photo editing. Changing a text or layout prompt generates a new image rather than editing source pixels.
3
Does it render English and Chinese text perfectly?
MOSS-Image supports both languages and has demonstrated detailed lettering. Results still need review for spelling, punctuation, reading order and typography; describe exact wording and line breaks when they matter.
4
How does JSON layout control work?
The model family accepts a JSON layout through its text-conditioning interface. A layout describes the scene and individual text or objects, with bounding boxes controlling their intended positions. It is experimental and does not guarantee exact placement.
5
Which image sizes and formats are configured here?
The generator is configured for a 1024-pixel long edge with five ratios: 1:1, 4:3, 3:4, 16:9 and 9:16. Its output formats are JPEG, PNG and WebP.
6
How should I write a MOSS-Image prompt?
Describe what the completed image should look like: subject, setting, style, lighting and composition. Specify literal text and its hierarchy. English is the recommended prompt language, although the model also supports other languages.
7
Can MOSS-Image create video or editable design files?
Its native output is a raster image. It does not generate video or guarantee editable text layers, SVG artwork or layered design files.
8
Where can I use MOSS-Image?
You can use the MOSS-Image AI Image Generator on MOSS Image.