Your app needs AI media.
You send one request.
One API, same for every model — Mynth translates it to what the provider needs.
Mynth validates it and locks a fixed price.
A background job takes over.
Inputs are resized and classified for the model.
Your prompt is rewritten with curated style tags.
Routed to the fastest available provider.
Failures retry across providers. You never see them.
Your media is generated.
The result is content-rated and formatted.
Delivered to the CDN, your storage, your webhooks.

content ratinginputsdestination

const task = await mynth.image.generate({model: "openai/gpt-image-2",size: "auto",prompt: "our product, studio light",rate: true,inputs: ["https://acme.dev/refs/product.png","https://acme.dev/refs/style.png",],});

{images: [{url: "https://cdn.mynth.io/m/img_8f3kp2.png",mynthUrl: "https://cdn.mynth.io/m/img_8f3kp2.png",rate: { level: "sfw" },...rest,},],...rest,}

03the toolkit

Built for real work.

Every feature exists because we needed it ourselves. No bloat — just the tools that make image generation in production painless.

01

Unified API

One payload for every model. Switch providers without touching integration code — unsupported parameters are adapted automatically.

02

Fully-typed SDK

Model IDs, parameters, and responses are all typed. Autocomplete that actually knows which sizes FLUX accepts.

03

Magic Prompts

Leave it on for every request. Good prompts pass through untouched; rough ones get rewritten and tuned to the model — behind one flag.

04

Content Rating

Rate any image against built-in or custom scales. Enforce your app's content policy without running your own classifier.

05

Public Access Tokens

Scoped, single-use tokens your frontend can use to poll task status directly. No background workers, no proxy endpoints.

06

Webhooks

Register a URL and Mynth calls you when the generation completes. Your backend never polls.

07

Async & Sync Modes

Synchronous responses for quick results, async with webhooks and polling for production workloads. Same payload either way.

08

Auto Size Selection

Pass size: "auto" and Mynth picks optimal dimensions for the prompt and model. One less decision per request.

04magic prompts

The prompt that adjusts itself.

Magic Prompt is a single flag you turn on for every request. Send a clean, model-ready prompt and it proxies straight through, untouched. Send something rough — another language, typos, a paragraph of story with one image buried inside — and Mynth rewrites it into a prompt the model can actually use, tuned to the model you picked.

You never have to decide which case you're in. It's the prompt-engineering layer you'd otherwise build, branch, and maintain yourself — replaced by a boolean.

magic_prompt:true

one example — when it has work to do

you send

Create a full-body shot of the woman described below:

"maren had spent thirty winters on the northern coast. her copper hair was going grey at the temples now, and a pale scar ran from her cheekbone to her jaw — a souvenir from the siege at Hollow Bay. she pulled her salt-stained wool cloak tight against the wind..."

▼ magic_prompt: true

mynth sends to the model

Full-body portrait of a weathered woman in her fifties, long copper hair greying at the temples, a pale scar running from cheekbone to jaw, wearing a salt-stained grey wool cloak, standing on a rugged windswept northern coastline, overcast light, photorealistic, sharp detail

Pointed at a tag-based model like Illustrious? Magic Prompt returns weighted tags and model-specific keywords instead — same flag, no extra work.

Already good? Untouched.

A clean, model-ready prompt proxies straight through. You pay no tax for getting it right.

Messy content, handled

Wrap a story, a product description, or raw LLM output in an instruction. Magic Prompt finds the image you actually want — typos and all.

Any language

Prompt in German, Japanese, or anything else. Mynth translates to English before the model ever sees it.

Model-aware phrasing

Natural language for SOTA models. Tag-based prompting and model-specific keywords for SDXL families like Illustrious.

06output

Made with one payload.

Every image below was generated through the Mynth API — different models, identical integration. Hover for the model behind each one.

Grok Imagine
autumn portrait, red hair against turning maple leaves

FLUX.2 [max]
bioluminescent leopard, glowing mushrooms, dark jungle

Gemini 3 Pro Image
the word spelled out entirely in tropical fruit, studio light

Recraft V4
low angle portrait in tall grass, windswept, overcast sky

Qwen Image 2.0
dragon embroidery on dark silk, obsidian sphere, macro detail

Seedream 5.0 Lite
floating crystal citadel, dragons circling at dawn

Bismuth Illustrious Mix
portrait, warm rim light, dark background, best quality

FLUX.2 [pro]
golden hour portrait, shadows through window blinds

Z-Image Turbo
film still, woman at the shore at dusk

Illustrious
epic wallpaper, ancient forest, dark muted greens

Generated with Recraft V4 Pro — Recraft V4 Pro

Generated with Gemini 3.1 Flash Image — Gemini 3.1 Flash Image

Generated with GPT Image 2 — GPT Image 2

Generated with Grok Imagine — Grok Imagine

Generated with Qwen Image 2.0 Pro — Qwen Image 2.0 Pro

Generated with Seedream 5.0 Lite — Seedream 5.0 Lite

Generated with Grok Imagine Image Quality — Grok Imagine Image Quality

Generated with Krea 2 Medium — Krea 2 Medium

Generated with Krea 2 Large — Krea 2 Large

05content safety api

Rate any image.
Define your own scale.

POST /image/rate accepts up to 10 image URLs and returns a content rating for each, in the same response. Use the built-in sfw / nsfw scale or define 2–7 custom levels that map exactly to your app's content policy.

Batch up to 10 images per request
Custom rating scales with natural-language descriptions
Per-image error handling — one failure doesn't block the rest
Rate images from any source, not just Mynth generations

pricing

$0.0002/ image rated

5,000 images per $1 — flat rate, no minimum

POST/image/rate

custom scale — 2 to 7 levels

GSuitable for all audiences

PGMild content, general audiences

PG-13Mature themes, not for children

RRestricted, adult themes

XExplicit adult content

response

// results[]

hero-banner.webpg

action-scene.webppg13

product-photo.webpg

07developer experience

Zero to image in three steps.

Install

01

$ bun add @mynthio/sdk

One dependency. Fully typed, zero config.

Generate

02

const task = await mynth.image.generate({prompt: "a fox in a cyberpunk city",});

Model, size, and prompt enhancement are handled automatically.

Receive

03

webhook: { url: "https://api.you.com/hook" }// or poll from the frontend with a// scoped public access token

No queues, no workers, no infrastructure to babysit.

See the full SDK documentation

mynth

API for AI media

Built for real work.

Unified API

Fully-typed SDK

Magic Prompts

Content Rating

Public Access Tokens

Webhooks

Async & Sync Modes

Auto Size Selection

The prompt that adjusts itself.

Already good? Untouched.

Messy content, handled

Any language

Model-aware phrasing

Made with one payload.

Rate any image.
Define your own scale.

Zero to image in three steps.

Install

Generate

Receive

Ship media generation this afternoon.

API for AI media

Built for real work.

Unified API

Fully-typed SDK

Magic Prompts

Content Rating

Public Access Tokens

Webhooks

Async & Sync Modes

Auto Size Selection

The prompt that adjusts itself.

Already good? Untouched.

Messy content, handled

Any language

Model-aware phrasing

Made with one payload.

Rate any image.Define your own scale.

Zero to image in three steps.

Install

Generate

Receive

Ship media generation this afternoon.

Rate any image.
Define your own scale.