Designers
Explore posters, title graphics, and layout concepts that combine lettering with imagery. Use structured prompts to communicate where key elements should appear.
MOSS Image AI Image Generator
Create images with Chinese and English lettering, detailed scenes, and a choice of photographic or illustrated styles. MOSS Image also accepts JSON layouts to guide where text and objects appear.
Bilingual Text
Layout Control
Long Prompts
Multiple Styles
Preparing the editor…
What Is MOSS Image?
An image model that brings words, visual detail, and composition into one prompt.
MOSS Image, officially named MOSS-Image, is a text-to-image model family from OpenMOSS, released on October 16, 2026. Its 5-billion-parameter backbone supports prompts up to 4K tokens and can generate both photographic and illustrated images.
The family includes MOSS-Image-Base and MOSS-Image-SFT. SFT is fine-tuned from Base on curated data; both use the same architecture and prompt interface. They are two checkpoints from the same release, rather than successive model generations.
Choose how to describe the image you want to create:
- Natural-language prompts: Describe the subject, scene, lighting, style, and any lettering in ordinary sentences.
- Tag sequences: Use concise tags to describe visual style and image content.
- Structured layouts: Describe text and objects in JSON, with bounding boxes to guide their arrangement.
Layout prompts are experimental. Review generated text and placement before using an image.
Changing a layout generates a new composition; it does not establish support for editing an uploaded photo.
MOSS Image Examples
Six published OpenMOSS examples, from handwritten text to illustration and panoramic scenes.
MOSS Image
Handwritten Words in an Illustrated Classroom
MOSS Image
A Reflection in the Rain
MOSS Image
Bold Shapes and Primary Colors
MOSS Image
A Marble Figure with Sculpted Detail
MOSS Image
Pencil Lines on a Printed Worksheet
MOSS Image
A Panoramic Shanghai Skyline
Key Features of MOSS Image
Explore six capabilities that help turn a visual idea into a detailed image.
Chinese and English Lettering
Include words directly in an image, from notes and signs to longer passages. Describe the exact wording and intended line breaks, then check the result for spelling and omissions.
Native JSON Layout Control
Guide the placement of text and objects with descriptions and bounding boxes. This experimental input format expresses your intended composition, without guaranteeing pixel-perfect placement.
Room for Detailed Prompts
Prompts can contain up to 4K tokens, giving you room to describe multiple subjects, materials, lighting, and their relationships within a scene.
Photographic and Illustrated Styles
Explore realistic-looking scenes, anime, sketches, and graphic artwork. The published examples show different visual styles within the same model family.
Flexible Image Proportions
Choose square, portrait, or landscape images on Picir AI with five aspect ratios: 1:1, 4:3, 3:4, 16:9, and 9:16. Output uses a fixed 1K setting, with JPEG, PNG, or WebP formats.
Multiple Ways to Prompt
Write a natural-language description, use tags, or combine prose with a structured layout. You can start simply and add more explicit composition guidance when needed.
Who Is MOSS Image For?
Use text, illustration, and composition controls for a wide range of visual projects.
Illustrators and Storytellers
Develop characters, atmospheric scenes, and illustrated story ideas. Describe the mood, materials, and visual relationships in a detailed prompt.
Content Creators
Create announcement graphics, bilingual visuals, and image concepts for posts or presentations. Review every generated headline before sharing it.
Educators and Writers
Explore classroom scenes, handwritten notes, and text-bearing illustrations for explanatory content. Check both the wording and factual accuracy of the finished image.
How to Use MOSS Image on Picir AI
Describe your idea, choose your image settings, and generate with MOSS Image.
1
Describe the Image
Write the subject, setting, visual style, and desired lettering. English is recommended in the official guide; Chinese prompts are also supported.
2
Choose Your Image Settings
Select an aspect ratio and JPEG, PNG, or WebP output. MOSS Image uses a fixed 1K setting; choose how many images you want to generate.
3
Generate and Refine
Generate your images, then review the results in your workspace. Check lettering and composition, and refine your prompt for another version.
MOSS Image Specifications
Model information and image-generation settings available on Picir AI.
| Developer | OpenMOSS Team |
| Release date | October 16, 2026 |
| Generation mode | Text to image |
| Model architecture | 5B-parameter diffusion transformer |
| Prompt formats | Natural language, tag sequences, and JSON layouts |
| Prompt input on Picir AI | Up to 5,000 characters |
| Text rendering | Chinese and English |
| Output resolution | 1K (fixed) |
| Aspect ratios | 1:1, 4:3, 3:4, 16:9, 9:16 |
| Output formats | JPEG, PNG, WebP |
| Layout control | Experimental text and object placement through JSON prompts |
| Reference images | No image upload; describe your image in the prompt |
Last updated: October 21, 2026
MOSS-Image-Base vs MOSS-Image-SFT
Compare the two checkpoints released together in the MOSS Image family.
| Specification | MOSS-Image-Base | MOSS-Image-SFT |
|---|---|---|
| Release date | October 16, 2026 | October 16, 2026 |
| Checkpoint role | Foundational model | Fine-tuned from Base on curated data |
| Backbone | 5B parameters | Same 5B architecture |
| Prompt formats | Natural language, tags, JSON layouts | Natural language, tags, JSON layouts |
| Prompt window | Up to 4K tokens | Same family interface |
| Text rendering | Chinese and English | Chinese and English |
| Layout control | Experimental | Shared layout interface |
| Official whiteboard example | Less organized text grouping | Clearer grouping and more organized layout |
MOSS-Image-Base Advantages: Base preserves a foundational checkpoint for adaptation and custom fine-tuning, including LoRA training.
MOSS-Image-SFT Advantages: SFT adds curated-data fine-tuning and shows clearer text grouping in the official whiteboard comparison.
Consider SFT when evaluating ready-to-use image quality, and Base when planning adaptation or fine-tuning. Compare both on your own prompts: one published example does not establish a universal winner.
MOSS Image Pricing
Choose a Picir AI plan and use credits to generate images with MOSS Image. See the current cost per image below.
Basic
$10
Billed Annually$15
USD/month
- 600 credits/month
- All features available
- Unlimited downloads per day
- Assets owned by customer
- Faster generation speed
- Priority support
Pro
Flash Sale 50%
$14.5
Billed Annually$29
USD/month
- 1600 credits/month
- All features available
- Unlimited downloads per day
- Assets owned by customer
- Faster generation speed
- Priority support
Max
$49
Billed Annually$59
USD/month
- 4000 credits/month
- All features available
- Unlimited downloads per day
- Assets owned by customer
- Faster generation speed
- Priority support
Prime
$99
Billed Annually$119
USD/month
- 8000 credits/month
- Team management
- Up to 5 team members
- All features available
- Unlimited downloads per day
- Assets owned by customer
- Faster generation speed
- Priority support
MOSS Image uses a fixed credit cost per generated image. The examples below show the cost for one image at the fixed 1K setting.
| Image settings | Credits per image |
|---|---|
| Square · 1:1 · JPEG · 1K | 8 |
| Landscape · 16:9 · PNG · 1K | 8 |
| Portrait · 9:16 · WebP · 1K | 8 |
Aspect ratio and output format do not change the per-image cost. Generating multiple images charges separately for each image.
FAQs
MOSS Image Questions and Answers
Find answers about prompts, layouts, model versions, and image-generation limits.
1
What is MOSS Image?
MOSS Image is an OpenMOSS text-to-image model family with a 5B backbone. Released on October 16, 2026, it includes Base and SFT checkpoints and supports photographic and illustrated outputs.
2
Can MOSS Image render Chinese and English text?
Yes. Both are documented capabilities, with published examples of handwriting and text within scenes. Exact spelling, line breaks, and completeness still require inspection.
3
Does JSON layout control guarantee exact placement?
No. The official guide describes layout prompts as experimental. Bounding boxes guide composition, so text or objects may differ from the requested position or size.
4
Can MOSS Image edit an uploaded image?
MOSS Image on Picir AI generates new images from text and does not accept uploaded reference images. Changing a JSON layout creates a new composition rather than editing an existing photo.
5
What image sizes does MOSS Image support?
On Picir AI, MOSS Image uses a fixed 1K setting and supports 1:1, 4:3, 3:4, 16:9, and 9:16 aspect ratios. You can choose JPEG, PNG, or WebP output. Resolution is fixed rather than a selectable control.
6
How should I write a prompt?
Describe the subject, setting, style, lighting, and exact text you want to see. The official guide recommends English prompts and suggests explicit line breaks for lettering. Natural language, tags, and JSON layouts can also be combined.
7
What is the difference between Base and SFT?
SFT is fine-tuned from Base on curated data. Both share the same architecture and prompt interface; SFT is not a larger or later-generation model. The official whiteboard example shows clearer text grouping with SFT.
8
Where can I use MOSS Image?
You can use the MOSS Image AI Image Generator on Picir AI. Enter a prompt, choose your image settings, and generate your images directly from this page.