# MOSS Image API (OpenMOSS)

> MOSS Image from OpenMOSS is an open-weights 5B text-to-image model that renders legible Chinese and English text and places text and objects where a JSON layout says, at 1K, 1.5K or 2K.

Page: https://cheapergen.com/models/openmoss/moss-image · API reference: https://cheapergen.com/models/openmoss/moss-image/api · Live status and uptime: https://cheapergen.com/models/openmoss/moss-image/status

## Pricing
Struck-through price: fal API: fal's public price for MOSS Image, $0.05 per megapixel, on the output's pixels (OpenMOSS sells no international API) — https://fal.ai/models/fal-ai/moss-image. Prices verified October 22, 2026.

| Tier | fal API | CheaperGen | Save |
|---|---|---|---|
| 1K | $0.0500/image | $0.0384/image | 23% |
| 1.5K | $0.1125/image | $0.0826/image | 27% |
| 2K | $0.2000/image | $0.1534/image | 23% |

## Specs
- Resolutions: 1K (1024 × 1024) · 1.5K (1536 × 1536) · 2K (2048 × 2048)
- Aspect ratios: 1:1 · 4:5 · 5:4 · 3:4 · 4:3 · 2:3 · 3:2 · 9:16 · 16:9 · 1:2 · 2:1 · 9:21 · 21:9 · 1:3 · 3:1
- Canvas at 16:9: 1K 1360 × 768 · 1.5K 2048 × 1152 · 2K 2720 × 1536
- Prompt input: Natural language, tags or JSON layout
- Model: 5B DiT, Qwen3.5 text encoder
- Weights: Open (Apache-2.0)
- Enable Safety Checker: Default true · can be set to false over the API

## Endpoints

### Text to Image — `POST https://cheapergen.com/api/v1/run/openmoss/moss-image/text-to-image`

| Param | Type | Required | Description |
|---|---|---|---|
| prompt | string | yes | Natural language, comma-separated tags or a JSON layout (a scene caption plus items with text and bounding boxes, in pixels from the canvas centre). Put the exact words to render in quotes. |
| aspect_ratio | string (enum) (1:1, 4:5, 5:4, 3:4, 4:3, 2:3, 3:2, 9:16, 16:9, 1:2, 2:1, 9:21, 21:9, 1:3, 3:1) | no | Output image shape. At 1K: 1:1 = 1024 × 1024, 4:5 = 912 × 1136, 5:4 = 1136 × 912, 3:4 = 880 × 1168, 4:3 = 1168 × 880, 2:3 = 832 × 1248, 3:2 = 1248 × 832, 9:16 = 768 × 1360, 16:9 = 1360 × 768, 1:2 = 720 × 1440, 2:1 = 1440 × 720, 9:21 = 656 × 1552, 21:9 = 1552 × 656, 1:3 = 576 × 1760, 3:1 = 1760 × 576. |
| resolution | string (enum) (1K, 1.5K, 2K) | no | Output size, the same price at every aspect ratio: 1K = 1 MP (1024 × 1024), 1.5K ≈ 2.25 MP (1536 × 1536), 2K = 4 MP (2048 × 2048) at 1:1, with 1 MP = 1024² px. Other ratios keep the pixels: each side is the 1:1 side × √ratio, rounded down to a multiple of 16 (e.g. 2K 16:9 = 2720 × 1536). |
| seed | integer | no | Use the same seed and settings to reproduce a result. Leave empty for random. |
| enable_safety_checker | boolean | no | If set to true, the safety checker will be enabled. Content our Acceptable Use Policy prohibits stays blocked either way. |

```bash
curl -X POST https://cheapergen.com/api/v1/run/openmoss/moss-image/text-to-image \
  -H "Authorization: Bearer $CHEAPERGEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "prompt": "A hand-painted wooden sign outside a tea shop that reads “山间茶舍 MOUNTAIN TEA HOUSE — OPEN DAILY 9–6”, morning mist, warm light",
  "aspect_ratio": "1:1",
  "resolution": "1K"
}'
```

## Getting the result
Jobs are async. Poll `GET https://cheapergen.com/api/v1/generations/{id}` until status is succeeded/failed/canceled, or pass `webhook_url`. Failed generations are refunded automatically.

## FAQ
**How is MOSS Image priced?**
Per image by resolution — 1K, 1.5K or 2K — at the same price for every aspect ratio. Failed generations are not charged.

**How do I place text with a JSON layout?**
Send the layout JSON as the prompt. Bounding boxes are pixel coordinates from the canvas centre, so use the canvas of the resolution and aspect ratio you request (listed in their descriptions and in the specs).

**Is this OpenMOSS's model?**
Yes. MOSS Image is OpenMOSS's open-weights model, served on GPU inference for open-source models.

**Can I use the output commercially?**
Yes — OpenMOSS releases the weights under Apache-2.0 — subject to our Terms of Service.

MOSS Image is OpenMOSS' open-weights model. CheaperGen provides API access to the open weights, run on GPU inference for open-source models, and does not own the model.