Short answer: to use MOSS Image AI, describe the picture you want and put every word that must appear in quotes — a poster with the headline "SUNNY COVE", a menu with "Flat White $4" — then generate. You can do it in the browser with the MOSS Image AI Generator (no GPU, no install; new accounts get free credits), or run the open model yourself in ComfyUI if you have a strong graphics card. Either way, proofread the text before you use the image.
What is MOSS-Image?
MOSS-Image is an open-source text-to-image model released by the OpenMOSS team on 16 October 2026: a 5-billion-parameter diffusion transformer under the Apache 2.0 licence. Its selling point is text inside images — headlines, menus, labels, signs, in English and Chinese — plus control over where things go in the frame. That is exactly what most image models get wrong, and why it trended within days.
What it is good for
- Posters and flyers: a big headline, a date line, an illustration.
- Menus and price boards: several lines, each with a price aligned on the right.
- Product labels and packaging mock-ups: a brand name and a small line of details.
- Infographics: a title and a few numbered steps with short captions.
- Social graphics in Chinese or English: festival posters, announcements, quotes.
You can see real, unedited outputs of all five, with the prompts that made them, on the MOSS Image AI Generator page.
Three prompt templates you can copy
Change the words in quotes and the style; keep the structure.
Poster
A retro travel poster for a seaside town. Big bold headline at the top: "SUNNY COVE".
Below it, smaller text: "Summer Jazz Festival · July 12–14". Illustrated lighthouse,
turquoise sea, warm sunset colors, flat vector style, plenty of clean space around the text.
Menu
A cafe menu board on a dark green chalkboard, hand-lettered white chalk. Title at the top:
"MORNING MENU". Three lines below, each with a price on the right: "Flat White $4",
"Avocado Toast $9", "Blueberry Muffin $3". Small coffee cup doodles in the corners.
Infographic
A clean flat infographic on a white background titled "HOW TO BREW TEA". Three numbered
steps in a vertical row, each with a simple icon and a short caption: "1. Boil water",
"2. Steep 3 minutes", "3. Enjoy". Pastel green and blue palette, modern sans-serif type.
Rules that make the difference:
- Quotes around every word that must appear. Anything not in quotes is a description, not text.
- Say where it goes: "at the top", "under it", "price on the right", "along the bottom".
- Keep each piece short. Titles and labels are reliable; full paragraphs are where letters slip.
- Name the lettering: bold sans-serif, elegant serif, hand-lettered chalk, neon.
- Leave space: "plenty of clean space around the text" keeps words off busy backgrounds.
How layout control works
Beyond plain prompts, MOSS-Image can take a structured layout: a list of elements (a headline, a price, an image area) with the box each one should occupy. That is useful when you need a precise arrangement — a label with fixed zones, a poster series that must line up.
On the online generator you write plain prompts for now. Describing positions in words, as in the templates above, is enough for most posters and menus: the model plans the layout from your description before it draws.
Running MOSS-Image in ComfyUI or locally
If you want to run it yourself:
- Official code and weights: OpenMOSS/MOSS-Image on GitHub.
- ComfyUI: there is an official workflow in the ComfyUI documentation.
- Hardware: a 5B model needs a graphics card with plenty of video memory; check the repository for the current requirements and smaller variants.
It is worth it if you generate in bulk, want to fine-tune on your own style, or need everything offline. To try the model, make a few designs or compare it with others, the online generator is quicker — nothing to download.
Where MOSS-Image still falls short
- Long text: a full paragraph can pick up a wrong letter or a repeated word.
- Tiny type: fine print and footnotes blur at about one megapixel.
- Exact fonts and logos: it imitates a style; it does not reproduce a specific typeface or trademark.
- Speed: roughly 20 to 90 seconds per run online — slower than the fastest image models.
Always read every word before you print, post or send an image.
MOSS-Image or another image model?
We only compare what we run. On Vuddex, MOSS Image is the choice whenever the words in the picture matter. MiniMax image-01, also on our AI Image Generator, is the better pick for pictures without text: it costs a fraction of the credits, comes back in seconds and can keep a real person’s face from a reference photo. We have not run a side-by-side test against other text-rendering models, so we won’t claim a winner.
FAQ
Is MOSS Image free? The model is open (Apache 2.0) and OpenMOSS hosts a free demo on Hugging Face, which can have a queue. On Vuddex each image costs credits; new accounts get free credits to try it, and failed images are refunded.
Can it write Chinese? Yes — short Chinese titles and slogans come out reliably. Put the Chinese text in quotes and check long passages carefully.
Is Vuddex the official MOSS-Image site? No. Vuddex is independent and runs the open model through a hosting provider.
Can I use the images commercially? The model licence allows it; on Vuddex, commercial use requires a paid plan, as in our terms.
Try it now with the MOSS Image AI Generator.