Coming soon on LazyKiwi
MOSS Image
An open text-to-image model built for images with words in them: posters, covers and layouts where the text has to come out clean. MOSS Image is not in LazyKiwi yet; Qwen Image, another open model that handles in-image text, is ready in the image generator now.
Using it
How to use MOSS Image
MOSS Image runs outside LazyKiwi today.
Run it locally
Download MOSS-Image-SFT from Hugging Face or ModelScope and run it with the repository's pipeline, the local Gradio app or ComfyUI.
Use the hosted demo or API
The README links a Hugging Face Space demo and the MOSI API platform (available from October 20, 2026).
Need an image with text now?
LazyKiwi's image generator runs Qwen Image, GPT Image 2.5 and Z-Image in the browser with free credits on sign-up.
What it is
What MOSS Image is built for
The OpenMOSS team (MOSI.AI) released the MOSS-Image weights on October 16, 2026, followed by a technical report on October 19 and access on the MOSI API platform on October 20. The repository describes it as "a 5B-parameter base DiT model with high-quality text rendering, layout generation, and efficient fine-tuning."
Text
Clear text inside the image
The model card leads with text rendering: "Generate clear, accurate text within images." The quick-start examples include both English and Chinese text.
- Readable in-image text
- English and Chinese examples
- Posters, covers, signs
Layout
Control where things go
MOSS-Image is pitched on layout generation: "Control the placement of text and visual elements," with prompts of up to 4K tokens, so a long brief describing a whole page can go in one prompt.
- Layout-aware generation
- Prompts up to 4K tokens
- Anime and photoreal styles
Open
Open weights under Apache 2.0
Both checkpoints, MOSS-Image-Base and MOSS-Image-SFT, are on Hugging Face; the model card states the weights and bundled inference code are released under the Apache License 2.0.
- Base and SFT checkpoints
- Apache 2.0 weights
- Hugging Face and ModelScope
Fine-tuning
Made to be adapted
"Easy to fine-tune" is one of the headline claims, and the team released a trainer (KohakuMossTrainer, October 18, 2026) for adapters on the base model.
- Base model for custom training
- Adapter trainer released
- ComfyUI support
Specs
MOSS Image specs at a glance
From the OpenMOSS GitHub repository and Hugging Face model cards, checked 2026-10-28. Items not stated there are marked "to be confirmed".
- Developer
- OpenMOSS team (MOSI.AI)
- Weights released
- October 16, 2026
- Architecture
- 5B-parameter DiT (diffusion transformer)
- Checkpoints
- MOSS-Image-Base, MOSS-Image-SFT
- Task
- Text to image
- Example output sizes
- 1024 × 1024 and 1456 × 1456 in the README
- License
- Apache 2.0 (weights and inference code)
- Where to run
- Hugging Face, ModelScope, ComfyUI, MOSI API platform
- Image editing
- To be confirmed
- In LazyKiwi
- Coming soon; Qwen Image is available now
Alternatives
MOSS Image vs models you can run in LazyKiwi today
Where each one fits. MOSS Image is not in LazyKiwi yet.
| Model | Strength | Best for | Where |
|---|---|---|---|
| MOSS Image | In-image text and layout | Posters, covers and pages with text | Coming soon |
| Qwen Image | Open model with in-image text and single-image edits | Text in images and quick edits today | LazyKiwi image generator |
| GPT Image 2.5 | Prompt-following and precise edits | Edits that must follow a detailed brief | LazyKiwi image generator |
| Z-Image | Fast open photoreal model | Fast photoreal drafts | LazyKiwi image generator |
FAQ
MOSS Image FAQ
What is MOSS Image?
MOSS-Image is an open-weight 5B-parameter text-to-image model from the OpenMOSS team (MOSI.AI), released on October 16, 2026. It focuses on clear in-image text, layout control and easy fine-tuning.
Is MOSS Image open source?
Yes. The Hugging Face model card says the weights and bundled inference code are released under the Apache License 2.0.
Can MOSS Image write text in images?
That is its main pitch: the README says it can "generate clear, accurate text within images," and its examples include English and Chinese text.
What is the difference between MOSS-Image-Base and MOSS-Image-SFT?
Base is the pretrained model meant for fine-tuning; SFT is the fine-tuned checkpoint meant for direct generation.
Can I use MOSS Image in LazyKiwi?
Not yet. It is coming soon; this page will say so when it is added. Qwen Image, GPT Image 2.5 and Z-Image are available in the image generator today.
Does MOSS Image edit photos?
The repository describes text-to-image generation; image editing is not listed as of October 28, 2026.
Make images with text today
MOSS Image is coming soon to LazyKiwi. Qwen Image is ready now in the same workbench, with free credits on sign-up.
More models
Image models you can run in LazyKiwi today.