AI content generation

Crevio turns prompts into production media and live web intelligence. Generate product images, marketing video, voiceovers, and run agentic web research — all through the API, billed per call against your credit balance. These operations are metered (they consume credits — see Usage & billing). Media operations are asynchronous: you kick off a job, then poll for the result. Finished media is saved to your MediaLibrary as file_… ids that you can attach directly to products, posts, and more.

How generation works

Most media calls don’t return the finished asset inline. They start a background Job and return a job you poll until it’s done.

Start a generation

Call POST /images/generate, POST /videos/generate, POST /audio/generate, or POST /audio/music. You get back a job.

Poll the job

Poll GET /jobs/{id} (or list with GET /jobs) until its status is terminal. The completed job references the resulting MediaLibrary file.

Use the file

The output lands in your MediaLibrary as a file_… id. Use it anywhere a file is accepted — e.g. a product’s images array.

Generated media always lands in the MediaLibrary first. You attach the file_… id — you never hand a raw CDN URL to a product or post.

Pricing

Pick a model with the model parameter; omit it to use the default. The simplest thing to pass is a quality tier — fast, balanced or quality — which resolves to the best current model at that price point for the endpoint you called. Tiers are chosen against independent blind-comparison rankings (Artificial Analysis arenas) per workload, and survive model churn, so prefer them unless you need a specific model. GET /images/models (and the video/audio equivalents) shows which id each tier currently resolves to.

{ "model": "balanced", "input": { "prompt": "a paper boat in a gutter" } }

Per-unit prices in USD. Note the unit differs per model — an image, a second of video, a megapixel or 1,000 characters. You are charged the quantity the provider actually bills, so a 10-second clip on a per-second model costs ten times its listed price, and a job that fails or is cancelled costs nothing.

Images

Tier	Model	Price
`fast`	`fal-ai/z-image/turbo`	$0.005 / megapixel
`balanced`	`fal-ai/nano-banana-2`	$0.08 / image — the default
`quality`	`openai/gpt-image-2.5/flare/text-to-image`	metered per request; highest-ranked text-to-image model, and the strongest at in-image text

Video

Tier	Model	Price
`fast`	`fal-ai/veo3.1/lite`	~$0.05 / second — the cheapest, with native audio; 4, 6 or 8-second clips
`balanced`	`minimax/h3-max/text-to-video`	~$0.05 / second — no audio
`quality`	`google/gemini-omni-flash/v1.1/text-to-video`	~$0.10 / second — native audio with lip-sync

These are what generations are actually billed per second of output; fal’s listed unit price is not always per second. GET /v1/videos/models returns each model’s current rate in credits as credits_per_second. fal-ai/veo3.1 (~$0.40 / second) is available but untiered: about four times the quality tier for similar rankings. Reach for it only when you need 4K output.

Audio (text-to-speech)

Tier	Model	Price
`fast`	`fal-ai/inworld-tts`	$0.01 / 1,000 chars
`balanced`	`fal-ai/elevenlabs/tts/turbo-v2.5`	$0.05 / 1,000 chars
`quality`	`alibaba/qwen-audio-3-tts`	$0.05 / 1,000 chars — highest-ranked voice on fal

fal-ai/elevenlabs/tts/multilingual-v2 ($0.10 / 1,000 chars) stays the default when you pass no model, and eleven-v3 is available for ElevenLabs’ most expressive delivery. Tiers switch voice families, so a voice value from one tier will not carry to another.

Music

Tier	Model	Price
`fast`	`minimax/music-3`	$0.002 / second of output
`balanced`	`fal-ai/lyria3/pro`	$0.08 / track — the default
`quality`	`mureka/api/generate/instrumental`	$0.225 / track — instrumental only

Tiers follow the Artificial Analysis instrumental arena. balanced and quality rank statistically level, so balanced is the default at about a third of the price.

These are the featured models, listed live with current pricing at GET /images/models, GET /videos/models, GET /audio/models, and GET /audio/music/models — not a closed set. Any fal.ai endpoint id is accepted, so you can pass a model that isn’t listed here (for example a newer video model) and it will be billed at that endpoint’s own rate. An endpoint fal doesn’t serve is rejected with invalid_model.

Images

POST /images/generate starts an image generation. The one thing that trips people up: the prompt goes under input: { prompt }, not at the top level. The input object is forwarded verbatim to the underlying model, so any model-specific fields go there too.

Image generation is the exception to Crevio’s “params at the top level” rule. The prompt must be nested under input. A top-level prompt is ignored.

cURL
SDK (TypeScript)
MCP (agent)

curl -X POST https://api.crevio.co/v1/images/generate \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "quality",
    "input": {
      "prompt": "bold brand hero image for a creative marketing agency, abstract gradient shapes and confident typography, studio lighting"
    }
  }'

import { Crevio } from "@crevio/sdk";

const crevio = new Crevio({ apiKeyAuth: "YOUR_API_TOKEN" });

const job = await crevio.images.generate({
  model: "quality",
  input: {
    prompt:
      "bold brand hero image for a creative marketing agency, abstract gradient shapes and confident typography, studio lighting",
  },
});

Generate a bold brand hero image for my marketing agency with abstract
gradient shapes and confident typography, and save it to my library.

The call returns a job. Poll it until it finishes:

curl https://api.crevio.co/v1/jobs/JOB_ID \
  -H "Authorization: Bearer YOUR_API_TOKEN"

When the job completes, the generated image is in your MediaLibrary as a file_… id.

Edit, upscale, and stock

Operation	What it does
`POST /images/edit`	Edit an existing image with a prompt (image-to-image)
`POST /images/upscale`	Increase resolution of an existing image
`GET /images/stock`	Search royalty-free stock imagery
`GET /images/models`	List featured image models, their tier, and live pricing

Video

POST /videos/generate follows the same shape family — pass a tier (or a model id from GET /videos/models), describe what you want, then poll for the result. Video has no default model, because it bills per second.

curl -X POST https://api.crevio.co/v1/videos/generate \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "balanced",
    "input": {
      "prompt": "energetic social banner clip for a personal fitness coach, dynamic motion graphics over a vibrant gradient backdrop"
    }
  }'

Poll GET /videos/{id} (or GET /jobs/{id}) until the render is done. Video can take a while — these are long-running jobs.

Audio (text-to-speech)

POST /audio/generate turns text into a voiceover. Pricing is per 1,000 characters, so pass the text you want spoken.

curl -X POST https://api.crevio.co/v1/audio/generate \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "balanced",
    "input": {
      "text": "Welcome to your free consultation. In this first call we will map out your goals."
    }
  }'

Poll GET /audio/{id} for the finished audio file.

Music

POST /audio/music makes an original track from a text description: a soundtrack, a background bed, a jingle or a song. Write prompt as a brief (genre, mood, tempo, instruments, energy) and say “instrumental, no vocals” for a bed under narration.

curl -X POST https://api.crevio.co/v1/audio/music \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "balanced",
    "input": {
      "prompt": "Hard-hitting country rock, driving drums, crunchy electric guitar, banjo lead, 140 BPM, instrumental, no vocals"
    }
  }'

Poll GET /audio/{id} (or GET /jobs/{id}); the track is result.audio. Tracks come back at the model’s own length, often a few minutes, so trim them to fit. Only the fast tier takes input.duration, an upper bound in seconds. The models take text only, so a reference recording can’t be passed. Describe its style instead.

Web tools

Beyond media, Crevio exposes a suite of live web operations — the same tools your AI Tasks use to research the open web. All are metered.

Endpoint	What it does
`POST /web/search`	Run a web search and return ranked results
`POST /web/read`	Fetch and clean a single URL into readable text
`POST /web/map`	Map the URLs/structure of a site
`POST /web/crawl`	Crawl a site and collect its pages
`POST /web/extract`	Pull structured data from one or more pages
`POST /web/research`	Agentic multi-step research with a synthesized answer

POST /web/research is the heavyweight: give it a query and a depth, and it fans out searches, reads sources, and synthesizes a cited answer.

curl -X POST https://api.crevio.co/v1/web/research \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "freelance web design pricing benchmarks 2026",
    "depth": "thorough"
  }'

The depth enum controls how hard it digs — at the cost of more credits and time:

`depth`	Behavior
`quick`	Fast, shallow pass (default)
`thorough`	Broader fan-out, more sources
`exhaustive`	Deep, multi-round research

Worked example: generate a product image and attach it

A common flow — create a draft product, generate a cover image, and attach it.

Create a draft product

curl -X POST https://api.crevio.co/v1/products \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{ "name": "Strategy Consulting Session", "status": "draft" }'

Returns the product (e.g. "id": "prod_abc123").

Generate the cover image

curl -X POST https://api.crevio.co/v1/images/generate \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {
      "prompt": "polished professional brand shot of a business consultant at a desk, warm office light, on a clean background"
    }
  }'

Returns a job — note its id.

Poll until the image is ready

curl https://api.crevio.co/v1/jobs/JOB_ID \
  -H "Authorization: Bearer YOUR_API_TOKEN"

When complete, the job references a MediaLibrary file (e.g. file_def456).

Attach the file to the product

curl -X PATCH https://api.crevio.co/v1/products/prod_abc123 \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{ "images": ["file_def456"] }'

The first id in images becomes the cover. An empty array clears the gallery.

Want the whole thing in one breath? Over MCP you can just ask: “Generate a cover image for my Strategy Consulting Session and set it as the product photo.”

Next steps

Launch a paid course

Chain products, pricing, and checkout end to end.

Tasks

Hand web research and content jobs to your AI workforce.

Usage & billing

How metered credits are consumed per call.

Conventions

The rules every endpoint follows.

Getting started

AI & agents

Core concepts

Integrations

Recipes

AI content generation

How generation works

Pricing

Images

Video

Audio (text-to-speech)

Music

Images

Edit, upscale, and stock

Video

Audio (text-to-speech)

Music

Web tools

Worked example: generate a product image and attach it

Next steps

Launch a paid course

Tasks

Usage & billing

Conventions

​How generation works

​Pricing

​Images

​Video

​Audio (text-to-speech)

​Music

​Images

​Edit, upscale, and stock

​Video

​Audio (text-to-speech)

​Music

​Web tools

​Worked example: generate a product image and attach it

​Next steps

Launch a paid course

Tasks

Usage & billing

Conventions

How generation works

Pricing

Images

Video

Audio (text-to-speech)

Music

Images

Edit, upscale, and stock

Video

Audio (text-to-speech)

Music

Web tools

Worked example: generate a product image and attach it

Next steps