Posters and flyers
A headline, a date and a paragraph of details on one poster: events, sales, openings, announcements.
MOSS Image on PicYep
Write the words you want in the picture — a headline, a whole menu, a paragraph in Chinese or English — and MOSS Image sets them into a poster, a sign or an infographic. PicYep is an independent studio that runs the model through a hosted API; it is not affiliated with the MOSS Image team.
- A set from one photo
- Plan and price first
- Not right? Free redo
- Free credits to start
Yep. That’s the shot.
New conversation
What is MOSS Image?
MOSS Image is an open-weight text-to-image model released on October 16, 2026 by MOSI.AI with Fudan University's OpenMOSS team, made for pictures that carry long passages of Chinese and English text. On PicYep it runs in the browser with nothing to install: describe the picture, put the exact words in quotation marks, and choose one of ten aspect ratios, 1K or 2K, and one to four images. A 1K image costs 25 credits.
New accounts start with free credits, and no card is needed to sign up. The cost of each photo is shown before you generate, and a request that ends without a photo gets its credits back. Your photos stay private to your account, carry no watermark and are yours to use for personal and commercial work.
What you can make with MOSS Image
Pictures where the words matter as much as the image.
Menus, signs and price lists
A full café menu with prices, a shop sign, a board of opening hours, in Chinese, English or both.
Infographics and slides
Steps, lists and short explanations laid out with headings: a how-to card, a recipe, a one-page summary.
How to use MOSS Image online
Describe the picture and its words
Say what the picture is, then write each piece of text in quotation marks, in the language it should appear in.
Pick the shape, size and count
MOSS Image is already chosen. Choose a tall shape for a poster or menu, a wide one for a banner, then 1K or 2K.
Generate and download
The cost in credits is on the button before you generate. Keep the version whose text reads best and download the original.
MOSS Image settings explained
What each choice under the prompt does when MOSS Image is the model.
| Setting | Details |
|---|---|
| Prompt | Up to 8,000 characters, room for a whole menu or several paragraphs. Before drawing, the model plans where each piece of text goes. |
| Text in the picture | Chinese and English, alone or mixed. Put the exact words in quotation marks; a line break in your text is a line break in the picture. |
| Aspect ratio | 1:1, 4:3, 3:4, 5:4, 4:5, 16:9, 9:16, 3:2, 2:3 or 21:9. Tall shapes suit posters and menus, wide ones banners and slides. |
| Resolution | 1K, about 1 megapixel (1024×1024 for a square), or 2K, about 2 megapixels (1440×1440, or 1920×1072 at 16:9). 2K gives small print more room. |
| Images per request | One to four. |
| Cost | 25 credits for a 1K image, 50 for 2K, shown on the button before you generate. A request that ends without an image gets its credits back. |
MOSS Image vs GPT Image, Nano Banana, Seedream and Qwen-Image
How the models compare on long text in pictures. Four of them run on PicYep, so you can give each the same prompt.
| Model | Made by | Open weights | Long text score | Known for | 1K image here |
|---|---|---|---|---|---|
| MOSS Image | MOSI.AI, OpenMOSS (Fudan) | Yes, Apache 2.0 | 96.3 | The highest long-text score of the open models; places text where a layout asks | 25 credits |
| GPT Image 2 | OpenAI | Not published | 99.0 | The top long-text score in the test; follows long prompts | 35 credits |
| Nano Banana Pro | Not published | 96.6 | Fine detail and complex instructions; also edits photos | 75 credits | |
| Seedream 5.0 Pro | ByteDance | Not published | 55.3 | A natural, photographic look | 35 credits |
| Qwen-Image 2512 | Alibaba Qwen | Yes | 64.9 | An earlier open model known for Chinese text | — |
Long text score: the mean of six English and Chinese tests, from short to long passages, on LongLongTextBench, the benchmark published with MOSS Image's technical report (October 2026); the share of the requested text that appears correctly. Qwen-Image does not run on PicYep. Prices here are live.
How to write long text into an AI image
Describe the picture first — a poster for a jazz night, a chalkboard café menu — then give every piece of text in quotation marks, in the order it should read: the headline, the date, the details. Say where each sits if it matters: "at the top", "in two columns", "small, along the bottom".
Write the description in English and keep Chinese text in Chinese; the two mix well in one picture. Break lines where you want them broken. For small print, choose 2K.
Open weights: run MOSS Image locally or online
MOSS Image is a 5-billion-parameter diffusion transformer trained from scratch, which reads prompts through a Qwen3.5-9B text encoder and draws through the FLUX.2 VAE. The weights are on Hugging Face in two versions: Base, for training your own LoRA, and SFT, tuned for finished pictures. It runs in ComfyUI, SGLang and vLLM-Omni.
Online on PicYep there is nothing to download and no graphics card or workflow to set up: it works on a phone, a tablet or a computer, and your images are kept in your account.
Can I use MOSS Image pictures commercially?
Yes. The weights and code are released under the Apache License 2.0, which allows commercial use with no limit on company size.
The images you make on PicYep are yours to use for personal and commercial work, with no watermark.
MOSS Image prompt tips
Quote every word
Put the exact text in quotation marks right after what it is: a headline reading "Grand Opening".
Break lines yourself
Line breaks are layout. Start a new line wherever the picture should, so a menu reads dish by dish.
Say where things go
Name the place of each block: the title at the top, prices on the right, the address small at the bottom.
Small print at 2K
The more words in the picture, the smaller they are. Draft at 1K, then make the keeper at 2K.
MOSS Image questions
Is MOSS Image free to use?
It is free to try on PicYep: a new account starts with free credits, and no card is needed to sign up. After that a 1K image costs 25 credits. The weights themselves are free to download under the Apache License 2.0.
Can MOSS Image write Chinese text?
Yes. It was trained to render both Chinese and English, and the two can share one picture, such as a bilingual menu or sign.
How much text can one picture hold?
The model reads prompts of up to about 4,000 tokens, and PicYep takes up to 8,000 characters — enough for a full menu or several paragraphs. The more words, the smaller each one is drawn, so choose 2K for long passages.
Who made MOSS Image?
MOSI.AI with researchers from Fudan University's OpenMOSS team, Shanghai Jiao Tong University and the Shanghai Innovation Institute. It was open-sourced on October 16, 2026, with a technical report on arXiv.
Can I change a photo with MOSS Image?
MOSS Image makes new pictures from words. To change a photo you already have, add it to the generator and pick FLUX 3 Klein, Nano Banana or GPT Image in the model menu.
Is PicYep made by the MOSS Image team?
No. PicYep is an independent AI image generator that runs MOSS Image through a hosted API, alongside other models such as GPT Image, Nano Banana and FLUX.
More to make.
Make a poster with MOSS Image
Describe the picture and quote its words — MOSS Image is already selected.
PicYep is an independent product. It is not affiliated with, endorsed by or officially connected to Black Forest Labs, Google or any other model maker named on this page. Model names belong to their owners.