Runs the real MOSS-Image · open model by OpenMOSS
MOSS Image AI generator
Turn a prompt into an image with MOSS-Image, the open 5B text-to-image model from OpenMOSS. Photoreal and anime styles, legible text inside the picture and prompts up to 4,000 characters. No GPU, no install.
6 free credits when you sign up · then from $0.04 an image · no subscription
MOSS-Image studio
0/4000
6 free credits when you sign up · no card needed
Not sure what to write? Start from one of these.
Gallery
Real MOSS-Image outputs, with their prompts
See the full gallery
Photography · Photo · 3:4
Portrait of an elderly fisherman mending a net on a wooden pier at dawn, weathered hands, soft mist over the harbour, 85mm lens, natural colour
Try this prompt
Anime · Anime · 16:9
Anime scene of a girl in a yellow raincoat waiting at a rural bus stop in summer rain, puddles reflecting blue hydrangeas, soft afternoon light
Try this prompt
Text & posters · Photo · 4:5
Hand-painted wooden sign above a mountain bakery door reading 'FRESH BREAD DAILY' in cream serif letters, pine trees behind, early morning sun
Try this prompt
Product · Product · 4:5
Amber glass perfume bottle standing on a slab of wet slate, a single white orchid petal beside it, soft rim light, deep green backdrop
Try this prompt
Text & posters · Poster & text · 2:3
Minimal travel poster for a fictional island called 'ISLA VERDE', bold sans-serif title across the top, stylised volcano and palm silhouettes, teal and coral palette
Try this prompt
Photography · Cinematic · 21:9
Lone cyclist crossing an empty desert highway at golden hour, long shadow on the asphalt, heat haze on the horizon, wide anamorphic framing
Try this prompt
Illustration & 3D · Illustration · 1:1
Cozy cutaway illustration of a treehouse library, children reading on floor cushions, paper lanterns, squirrels sorting books on the shelves, warm watercolour textures
Try this prompt
Illustration & 3D · 3D · 1:1
Cute 3D clay character of a mushroom chef holding a tiny wooden spoon, round cheeks, pastel kitchen background, soft studio lighting
Try this prompt
Why MOSS-Image
What MOSS-Image does well
The strengths the official model card describes, each one a control in the studio.
Photoreal and anime in one model
Trained for both photographic scenes and anime-style illustration, two looks most open models handle unevenly.
Legible text in the picture
Signs, labels, posters and packaging with the words you put in quotes, in English or Chinese.
Long, detailed prompts
Describe a whole scene in up to 4,000 characters. The model reads prompts of up to 4K tokens instead of cutting them short.
JSON layouts
Place a headline, a product and a background where you want them with a structured layout prompt.
Any frame, several takes
Nine aspect ratios from 1:1 to 21:9, Standard or HD, and up to 4 variations of every prompt.
An open model
Open weights under Apache 2.0. Use it here with no setup, or run it yourself when you outgrow us.
How it works
From prompt to image in three steps
No GPU, no install, no queue.
- 1
Describe it
Write what you want to see: subject, setting, light, framing. Or start from an example.
- 2
Pick a look
Choose a style, an aspect ratio, how many images and Standard or HD.
- 3
Download and share
Images arrive in seconds. Download them full size or share a link.
About the model
What is MOSS-Image?
MOSS-Image is a 5B text-to-image diffusion transformer released by OpenMOSS in October 2026, as a Base checkpoint for fine-tuning and an SFT checkpoint for everyday generation. A large language-model text encoder reads the prompt, which is why it follows long descriptions and renders text so well.
MOSS Image AI runs it on hosted GPUs so you can use it from any browser, with no 28 GB of weights to download and no ComfyUI to set up.
MOSS Image AI is an independent service. It is not affiliated with or endorsed by OpenMOSS. Every image it makes is AI-generated.
Read the full model guidePrompt tips
Write prompts MOSS-Image understands
MOSS-Image reads long, specific prompts well. Four habits get better images on the first try.
Read the full prompt guideLead with the subject
Say what the image is of first, then the setting, the light and the camera.
Quote the words
Text in quotes is rendered as written. Keep it short and say where it goes.
Choose the frame first
Pick the aspect ratio before you write: it changes how the scene is composed.
English or Chinese
Both work. English is the most reliable choice for everything except Chinese lettering.
Pricing
Simple credit packs
Buy once, no subscription. Credits never expire, and failed generations refund their credits automatically.
Starter
Enough to try every style
$0.06 per standard image
- 80 credits
- 80 standard or 40 HD images
- Credits never expire
- Images kept for 90 days
- Failed generations refunded
Creator
For regular projects
$0.05 per standard image
- 300 credits
- 300 standard or 150 HD images
- Credits never expire
- Images kept for 365 days
- Failed generations refunded
Studio
Best price per image
$0.04 per standard image
- 800 credits
- 800 standard or 400 HD images
- Credits never expire
- Images kept for 365 days
- Failed generations refunded
New here? Sign up and get 6 free credits first, no card needed.
Secure checkout by Stripe. Cards, Apple Pay and Google Pay.
FAQ
MOSS Image AI: questions, answered
What is MOSS-Image?+
MOSS-Image is an open-source text-to-image model released by OpenMOSS in October 2026: a 5-billion-parameter diffusion transformer trained for both photorealistic and anime-style images, with strong text rendering and support for long, structured prompts. This site runs it for you online.
Is MOSS Image AI free?+
New accounts get 6 free credits, enough for 6 standard images, with no card needed. After that you buy one-time credit packs; there is no subscription.
Do I need a GPU or ComfyUI?+
No. Everything runs on hosted GPUs, so a browser is all you need. If you'd rather run the open weights yourself, our ComfyUI guide shows how.
Is this really MOSS-Image?+
Yes. Every image is generated by the MOSS-Image model through a hosted inference provider; we don't swap in another model. The gallery shows real outputs with the exact prompts that made them.
How do I get legible text in an image?+
Put the exact words in quotes and say where they go, for example a sign reading 'OPEN 24H' above the door. Keep it to a line or two and check the spelling at full size. English and Chinese work best.
How long does an image take?+
Usually 10 to 30 seconds. HD images and batches of four take a little longer.
Can I use the images commercially?+
The model is released under Apache 2.0, and the images you generate are yours to use, including commercially. A model licence doesn't guarantee an image is free of third-party rights, so leave trademarks, real people and copyrighted characters out of work you sell.
What's the difference between Standard and HD?+
Standard images are about 1 megapixel (1024×1024 when square) and cost one credit. HD images are about 2.25 megapixels (1536×1536 when square) and cost 2 credits; use them for print and large banners.
Are my prompts and images private?+
Yes. Your images are visible only to you unless you create a share link, and we never use them to train models. You can delete them at any time.
What happens if a generation fails?+
The credits it used come straight back to your account. If only some images in a batch fail, you get back the credits for the missing ones.
Your next image is one prompt away
Sign up and get 6 free credits. No card, no GPU, no install.
Start creating