How to Use the MOSS-Image Turbo API

Disclosure: This article was written with AI assistance and reviewed by EmpirioLabs AI.

MOSS-Image Turbo is available on EmpirioLabs. It is OpenMOSS's text-to-image model for images that carry words: signs, posters, labels, menus, and layouts, in English and Chinese. It runs on POST /v1/images/generations, in the API and in the Playground.

What it does with text

One shot from a trail-sign prompt that asked for the words "SUMMIT 2.4 MILES". The same test against three other models is in MOSS-Image Turbo vs FLUX.2 Klein 4B vs Qwen Image 2.1 vs Seedream 5.0 Flash.

Your first request

curl https://api.empiriolabs.ai/v1/images/generations \
  -H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moss-image-turbo",
    "prompt": "A bakery storefront sign that reads \"FRESH BREAD DAILY\" in gold letters, morning light",
    "sync": true
  }'

With sync set, the response is the finished image: a signed URL to a 1024x1024 WebP, with the seed it used. Without it, the response is a job_id to poll at GET /v1/jobs/{job_id}, which suits long batches.

Sizes and options

• Aspect ratio. aspect_ratio takes 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, or 2:3.
• Exact size. size as WIDTHxHEIGHT, or width and height: each side a multiple of 16 from 512 to 2,048 pixels, and at most 1,048,576 pixels in total, such as 1344x768 or 2048x512.
• Several images. num_images takes up to 4. With a seed, the images use that seed and the ones after it (42, 43, and so on), and each image returns its seed.
• Formats. WebP by default, PNG or JPEG with format, and base64 instead of a URL with response_format set to b64_json.

Writing prompts with text

• Quote the words. Put the exact text in quotation marks, for example a sign that reads "OPEN 24 HOURS".
• Say how it looks. Name the lettering and the surface: hand-painted, gold serif, red neon, brush calligraphy, carved wood.
• Chinese works the same way. Put the characters in quotation marks, such as “春山茶舍”.

Operational notes

• Width and height are multiples of 16. A 1000x1000 request is refused with a 400 that says so; 1008x1008 and 1024x1024 work.
• Sampling is fixed. negative_prompt, num_inference_steps, and guidance_scale are refused with a 400 instead of being ignored.
• Text to image only. Image inputs and LoRA adapters are refused. To edit from a reference image, use a model with image editing such as FLUX.2 Klein 4B.
• Check posters for extra text. In one of our three test prompts it added a short line of lettering that was not asked for.

Pricing

Billed per image. A request for four images bills four, and a refused request is not billed. Generated images pass an automated content-safety check: an image flagged as explicit is not returned and is still billed. Current rates are on the model page and the pricing page.

Start building

Try MOSS-Image Turbo in the Playground, where every parameter is a live control, and read the API documentation for the full parameter reference.