MOSS-Image API
MOSS-Image — the Apache-2.0, 5-billion-parameter diffusion transformer from OpenMOSS — behind one REST call. The model to reach for when the image has to carry words: signs, posters, packaging, menus, slides, UI mock-ups and infographics, in English and Chinese, from prompts up to several thousand tokens.
- Price
- $0.006 / image
- Model id
- moss-image
- Sizes
- 7 presets
- Images per request
- 1–4
1024x1024, 832x1216, 1216x832, 896x1152, 1152x896, 1536x1024, 1024x1536
What it is good at
moss-image runs OpenMOSS’s MOSS-Image-SFT checkpoint with a 4-step Euler-native turbo adapter, so a 1 MP image takes about five seconds on a warm worker: a 5B diffusion transformer paired with a Qwen3.5-9B text encoder and the FLUX.2 VAE. The encoder reads prompts of up to about 4K tokens, so long, specific captions with exact copy in quotes come out as you wrote them — tall headline here, small italic tagline there — where SDXL-era models scramble the letters.
Use it for anything typographic or diagrammatic: posters and flyers, storefront signage, product packaging, menus, slides, charts, UI and app mock-ups, comics panels with speech bubbles. For photoreal people and scenes without text, krea-2 is faster and cheaper per image; for anime or comics styles, the krea-2-anime and krea-2-cartoon models.
- Text to image at seven supported sizes from 1 MP to 1.5 MP
- Legible text rendering in English and Chinese, including multi-line copy and numbers
- JSON layout prompts: captioned boxes for text, objects and characters, placed in pixels
- Bring your own MOSS-Image LoRAs by URL (up to 3 per request)
Generate an image
A minimal request needs a model and a prompt. Quote the exact copy you want rendered; describe the type style around it.
curl https://api.fastgencloud.com/v1/images/generations \
-H "Authorization: Bearer $FASTGEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "moss-image",
"prompt": "A vintage travel poster for Lisbon: a yellow tram climbing a steep cobbled street between pastel houses at golden hour, the title \"LISBOA\" across the top in tall cream serif letters, the tagline \"Ride the 28\" in smaller italic script along the bottom, flat colour print style with subtle paper grain.",
"size": "1024x1024",
"n": 1
}'Place text and objects with a JSON layout
MOSS-Image was trained on layout-annotated data, so a prompt that is a JSON layout object is read as one: a summary caption for the whole image plus a list of items, each with a label (text, object or character), a caption, the exact text for text items and a bounding box. Boxes are pixel edges with the origin at the canvas centre, in [top, left, bottom, right] order — on a 1024x1024 canvas the coordinates run from -512 to 512.
The prompt reaches the model verbatim, so send the JSON as the prompt string itself (no wrapper, no prose around it) and keep the width and height of the request equal to the canvas you laid the boxes out on.
{"summary":{"caption":"A clean product poster for a matte black espresso machine on a warm grey studio background"},
"items":[
{"label":"text","caption":"bold white sans-serif headline","text":"MORNING, UPGRADED","bbox":[-460,-440,-340,440]},
{"label":"object","caption":"the espresso machine, three-quarter view, soft rim light","bbox":[-300,-280,300,280]},
{"label":"text","caption":"small grey caption line","text":"Model E1 · ships in 48 h","bbox":[380,-300,440,300]}
],
"bbox_coord_space":"abs_centralized","bbox_format":"yxyx"}Parameters
- model
- moss-image
- prompt
- 1–2000 characters, passed to the model verbatim (plain sentences or a JSON layout)
- negative_prompt
- accepted but has no effect — the turbo recipe samples without classifier-free guidance
- size
- 1024x1024, 832x1216, 1216x832, 896x1152, 1152x896, 1536x1024, 1024x1536
- aspect_ratio
- alternative to size — picks the closest supported size
- n
- 1–4 images per request
- seed
- optional integer for reproducible output
- steps / guidance
- default 4 steps (6 at 1536x1024 and 1024x1536); steps 4–8, guidance fixed at 1
- loras
- optional [{ url, strength }] — public MOSS-Image LoRA files, up to 3
- webhook_url
- optional signed callback on completion
Unfiltered and uncensored
This endpoint applies no content filter. Your prompt, negative prompt, reference images and LoRAs reach the model verbatim, nothing is rewritten or injected, and outputs are not scanned or blocked. What your product allows is your call.
The one automatic rule is the minor-safety floor: a prompt that pairs a minor or age reference with sexual content is refused for that request only. Nothing else is filtered and the account is never touched by moderation.
How the API works
Every generation is a job. The POST returns 202 with a job id immediately, so your request thread is never blocked on a GPU. You then either poll GET /v1/jobs/{id} or register a webhook_url and receive a signed callback when the job finishes.
Send an Idempotency-Key header and a retried request replays the original job instead of charging you twice. Your wallet is debited when the job is submitted and refunded automatically if it fails.
curl https://api.fastgencloud.com/v1/jobs/$JOB_ID \
-H "Authorization: Bearer $FASTGEN_API_KEY"Pricing and billing
Pricing is prepaid and usage-based: you top the wallet up with a card and each job debits it. There is no subscription, no per-seat cost, and no monthly minimum.
Images above 1 megapixel are billed pro rata by pixel count — a 1.5 MP image costs 1.5x the listed rate. Failed jobs are refunded in full, automatically.
Frequently asked questions
- Is this the official OpenMOSS API?
- No. This is an independent host of the open-weight MOSS-Image model (Apache-2.0). OpenMOSS offers the model on its own MOSI platform; this endpoint runs the MOSS-Image-SFT checkpoint on the reference sampler, exposes the same REST contract as every other model here, bills per image from a prepaid balance, and takes your own MOSS-Image LoRAs by URL.
- How fast is it, and can I trade speed for quality?
- The endpoint runs the Euler-native 4-step turbo adapter for MOSS-Image-SFT, so a 1 MP image renders in about five seconds on a warm worker with the same text fidelity as the 25-step reference sampler at that size. For the 1.5 MP sizes or dense small type, raise steps to 6–8; the default already moves to 6 at 1536x1024 and 1024x1536. The 25-step classifier-free-guidance recipe is not exposed on this model id.
- Does it do image-to-image, inpainting or trained characters?
- Not on this model id: moss-image is text-to-image only. Image-to-image, inpainting and the characters API live on the Krea 2 lineup (krea-2, krea-2-anime, krea-2-cartoon), which keeps an SDXL companion checkpoint for exactly those jobs.
- Are my prompts rewritten or enhanced before they reach the model?
- No. The ComfyUI reference workflow ships an LLM prompt enhancer and a default negative prompt; this endpoint uses neither. Your prompt is encoded exactly as you send it, which is also what makes JSON layout prompts work.
- How long are generated images stored?
- Output URLs live for up to 7 days, then the files are deleted. Download anything you want to keep and store it yourself.
Also searched as
- MOSS-Image API
- MOSS Image API
- OpenMOSS image model API
- MOSS-Image text rendering
- MOSS-Image layout JSON
- hosted MOSS-Image
- MOSS-Image SFT API
- text to image with legible text API
- poster generation API
- infographic image generation API
- MOSS-Image pricing per image
- MOSS-Image LoRA API
Related
Start generating
Create an account, add a prepaid balance, and call the API with a key from the console. No subscription, no minimum, no per-seat pricing.
Get an API key