Gig poster
A screen-printed gig poster in orange and navy for a band called "THE LATE TRAINS", a night train crossing a bridge under a full moon, and the line "FRIDAY 9 PM · THE BLUE ROOM" along the bottom
Powered by MOSS Image from OpenMOSS · Independent, unofficial tool
MOSS Image AI Generator: Text to Image Model
MOSS Image is an open-source text-to-image model from the OpenMOSS team, out October 16, 2026 under the Apache 2.0 license. It's built for words inside the picture: posters, signs, sticky notes and menus, in English or Chinese. Here you type a prompt and get the image back for 1 credit, with no graphics card and no ComfyUI.
- What is MOSS Image?
- An open-source text-to-image model from OpenMOSS, out Oct 16, 2026.
- What is it good at?
- Words inside the picture, in English or Chinese, placed where you ask.
- Is it free?
- The weights are, commercial use included. Running them takes a graphics card.
- No GPU or ComfyUI?
- Use it here: 1 credit an image, 20 images for $11.90.
Size
about 1 megapixel · 1 credit
Your image will show up here.
What MOSS Image is
The OpenMOSS team put out two image models on October 16, 2026, code and weights together: MOSS-Image-Base, the plain model meant for people who train their own versions, and MOSS-Image-SFT, the same model fine-tuned on curated data.
Most image models treat lettering as an afterthought. MOSS Image puts it first: its makers list strong text rendering, prompts of up to 4K tokens, and a way to say where each piece of text and each object goes, as a JSON layout with boxes. It runs on a compact 5B backbone. ComfyUI, SGLang and vLLM-Omni supported it on release day, and within two days the community had a trainer that fits add-ons for it on a 6 GB graphics card.
Prompts that put words in the picture
Five to start from. "Use this prompt" puts it in the tool above with its shape; change the words in quotes to yours.
Sticky note
A yellow sticky note on a silver laptop lid, handwritten in blue pen: "Call mom Buy milk Finish the slides!"
Shop sign
A hand-painted wooden sign above a small bakery door that reads "Fresh Bread Daily", soft morning light, a bicycle leaning against the brick wall
Café menu
A café chalkboard menu titled "TODAY" in white chalk, with three lines underneath: "Iced latte 4.50" "Banana bread 3.00" "Soup of the day 6.00"
Sticker
A die-cut vinyl sticker of a cheerful cartoon cactus in a pink pot, with the words "STAY SHARP" in bold rounded letters on a banner below it, white border, plain grey background
Getting the words right
- Put the exact words in quotes, and say where they go: "across the top", "on the door".
- More than one line? Break the lines yourself, the way the sticky note and the menu above do. MOSS Image's makers say that's how to control where a line breaks.
- Write the rest of the prompt in English, even for Chinese text: it's the language most of the training data was in.
- Short lines come out cleaner than paragraphs. Check the spelling on the result before you print it.
MOSS Image vs FLUX 3 Klein vs Qwen Image 2.1
| MOSS Image | FLUX 3 Image Klein | Qwen Image 2.1 | |
|---|---|---|---|
| Made by | OpenMOSS (MOSI) | Black Forest Labs | Alibaba's Qwen team |
| Out | October 16, 2026 | October 15, 2026 | September 20, 2026 |
| License for the weights | Apache 2.0: commercial use allowed | FLUX Kommunity License | Research and evaluation only |
| What it does | Pictures from text, with words and layouts | Pictures from text, photo edits, ten references | Pictures from text and photo edits |
| On GetStepSnap | Yes, on this page | Yes, photo edits on /flux-3-klein | No (the license rules it out) |
More on the other two: FLUX 3 Klein edits your photo on this site, and Qwen Image 2.1's page explains its research-only license.
Is MOSS Image free?
The model is, and more freely than most: Apache 2.0 lets you use the weights for paid work, not just for research. The catch is the computer. OpenMOSS's quick start wants Python and an NVIDIA graphics card, and ComfyUI needs its newest version. If that's your thing, run it at home and pay nothing per image. OpenMOSS's demo on Hugging Face is free too.
Everyone else can use it here: no download, no setup, 1 credit an image.
What it costs
- 1 credit an image, in any of the five shapes. Packs cost $11.90 for 20 credits or $32.90 for 70, paid once: 20 or 70 images, about $0.47 each on the bigger pack. No subscription, and credits don't expire.
- If an image fails, its credit goes straight back to your balance.
- See credit packs
Questions
What is MOSS Image?
MOSS Image is a family of open-source text-to-image models from the OpenMOSS team, released October 16, 2026: MOSS-Image-Base and MOSS-Image-SFT, a version fine-tuned on curated data. Its makers list strong text rendering, control over where text and objects go, prompts of up to 4K tokens, anime and photorealistic styles, and a compact 5B backbone.
Is MOSS Image free?
The model is. Its weights are on Hugging Face and ModelScope under the Apache 2.0 license, which allows commercial use. OpenMOSS also runs a free demo on Hugging Face. Running it yourself takes an NVIDIA graphics card and some setting up. Here it isn't free: an image takes 1 credit, and packs cost $11.90 for 20 credits or $32.90 for 70, paid once.
Can I use MOSS Image without a GPU or ComfyUI?
Yes, three ways: on this page, where you type a prompt and get the image in your browser; in OpenMOSS's demo on Hugging Face; or through MOSI's own API platform, which opened on October 20. ComfyUI and a graphics card are only needed to run it on your own computer.
What is MOSS Image good at?
Words in the picture: a poster title, a shop sign, a handwritten note, a menu, in English or Chinese. Its makers built it to place text and objects where you say, and it takes long, detailed prompts. It also does anime and photo-style pictures without text.
How is MOSS Image different from FLUX and Qwen Image?
The license, mostly. MOSS Image's weights are Apache 2.0, so you can use them for paid work. FLUX 3 Image Klein's weights come under Black Forest Labs' FLUX Kommunity License, and Qwen Image 2.1's free weights are for research only. MOSS Image is aimed at text and layout; on GetStepSnap, FLUX 3 Klein edits photos and FLUX 3 Image makes pictures from a sentence.
Is this the official MOSS Image?
No. GetStepSnap is an independent service, not affiliated with or endorsed by OpenMOSS or MOSI. We run MOSS Image through an API provider, fal.ai, and charge credits for the images you make here.
Can I write Chinese text with it?
Yes. MOSS Image renders Chinese and English text. Its makers recommend writing the prompt itself in English, which most of its training data was in, and putting the Chinese words you want to see in quotes.
How big is the image?
About 1 megapixel, in five shapes: 864 by 1152 (3:4), 992 by 992 (1:1), 1152 by 864 (4:3), 736 by 1312 (9:16), 1312 by 736 (16:9). That's enough for a phone wallpaper, a social post or a print up to about 4 by 3 inches.
Can I use the images commercially?
We don't claim any rights to the images you make here: you may use them for personal or commercial purposes, subject to the law. The model's own license, Apache 2.0, also allows commercial use if you run it yourself.
Sources: the MOSS-Image project page, its GitHub README and user guide and MOSS-Image-SFT on Hugging Face, read October 21, 2026.
More AI photo tools
GetStepSnap is an independent product and is not affiliated with, endorsed by, or sponsored by OpenMOSS or MOSI. MOSS is their name for their models.