Azurade AI

One API key for Veo 3.1, Seedance and Nano Banana

Pay as you go for 45 image and video models, from a REST API or from an MCP server that Claude, ChatGPT, Cursor and Codex connect to with one sign-in. It's the credit balance you use in the web app, at the same prices.

MCP server
https://azurade.com/mcp
REST API
/api/v2, keys as Bearer tokens. Reference
Billing
A credit costs $0.02 to $0.04, no subscription. 4 free on signup.
Credits
Never expire. Replicate's prepaid credit lasts a year, fal.ai's 365 days (compared).
Failed generations
Refunded to the balance automatically, no ticket.

Watch Claude and Codex connect to Azurade in two clicks. The steps for every other client are below. Only want pictures in Claude? Read Claude image generation

Connect an MCP client

Claude (web and desktop)

  1. Open Customize > Connectors.
  2. Click +, then Add custom connector.
  3. Paste the URL.

On Team and Enterprise an owner adds it first under Organization settings > Connectors. The free plan allows one custom connector.

https://azurade.com/mcp

On first use the client opens an Azurade page where you sign in and click Allow, and it's connected without you copying a key.

A client that cannot sign in sends a key from your profile page instead:
claude mcp add --transport http azurade https://azurade.com/mcp --header "Authorization: Bearer sk-..."

Then ask for a picture in Claude:

A teal dragon breathing a stream of photos into the night sky, painterly

estimate_generation

model: "nano-banana-pro", resolution: "2K", aspect_ratio: "4:5"

← cost: 9, balance: 240, affordable: true

create_generation

same arguments

← status: "completed", cost: 9

outputs[0].url: https://azurade.com/api/files/output_result_files/r8f2k1x0q9m4b7d/dragon.png

A teal dragon breathing a stream of photos into a dark night sky
Here it is. That took 9 credits; 231 left.Made with Nano Banana Pro on Azurade

REST API

  1. Create a key on your profile page. GET /api/v2/models is public; everything else takes the key.
  2. POST /api/v2/generations/estimate takes the same body and returns the cost and whether your balance covers it. It starts nothing.
  3. The create call answers 202 with an id. Poll GET /api/v2/generations/<id> until the status is failed, or completed with files in outputs.

The MCP tools call the same code, so prices, refunds, content screening and error codes match. Download links need no authentication: anyone you share one with can open the file.

Full contract: the OpenAPI document, generated from the running code and browsable at /swagger/. Agents that read skills can start from SKILL.md.

Request
curl -X POST https://azurade.com/api/v2/generations \
  -H "Authorization: Bearer $AZURADE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "nano-banana", "prompt": "a lighthouse at dusk"}'
Response, then the poll
202 Accepted
{"id": "r8f2k1x0q9m4b7d", "model": "nano-banana", "type": "image",
 "status": "pending", "cost": 2}

GET /api/v2/generations/r8f2k1x0q9m4b7d
{
  "id": "r8f2k1x0q9m4b7d",
  "status": "completed",
  "cost": 2,
  "outputs": [{
    "url": "https://azurade.com/api/files/output_result_files/.../lighthouse.png",
    "content_type": "image/png"
  }]
}

What the agent can do

Eleven tools. A finished image comes back with its download link and a small inline preview, so the agent can look at what it made before deciding to try again.

Choose a model

list_models
Every model with its type and the credit price of a default request.
get_model
One model's parameters: names, defaults, allowed values.
estimate_generation
The exact credit cost of a request. Free, starts nothing.

Generate

create_generation
Starts an image or video and waits up to about 50 seconds for it.
get_generation
Status and file links; waits again while the job runs.
list_generations
Your history, newest first.
create_upload_url
A one-off link the agent's shell uploads a local file to with curl.

Prompts

enhance_prompt
Rewrites a short prompt using the target model's prompting guide.
moderate_prompt
Runs the content-policy screen on a prompt without generating.

Account

get_account
Your credit balance.
buy_credits
A card checkout link for a credit pack, for you to pay when the balance runs out.

Models and prices

Credits for a request that sends only a prompt, the figure estimate_generation returns for it. † marks a model whose price changes with resolution, duration or inputs; estimate those before you submit. A credit costs $0.02 to $0.04 depending on the pack or plan.

Image models 26

Model idBest forTakesCredits
nano-bananaQuick image edits and blending several reference imagestext, image2
nano-banana-2.1Images up to 4K with clean text and characters that stay the same across editstext, image2, price varies
qwen-image-2.1Generating and editing in one model, with transparent backgroundstext, image2, price varies
recraft-crisp-upscaleUpscaling an image with crisp edges and textimage2
remove-backgroundCutting the subject out onto a transparent backgroundimage2
seedream-5-flashFast, low-cost images and edits up to 2Ktext, image2
flux-devGeneral-purpose images with control over steps, guidance and seedtext, image3, price varies
gpt-image-2.5Photorealistic images with precise text up to 4K, and tighter editstext, image3, price varies
qwen-imageImages and edits with clean in-image text in Chinese or Englishtext, image3, price varies
qwen-image-editEditing one image with a text instructiontext, image3
qwen-image-edit-plusEdits that combine several input images, such as swapping outfitstext, image3
qwen-image-edit-plus-loraImage edits in the style of a custom LoRAtext, image3
seedream-3Photoreal images at up to 2048 pxtext3
seedream-4Up to 6 images per request, and edits with up to 10 referencestext, image3, price varies
seedream-4.5Detailed images up to 4K, and edits with up to 14 referencestext, image3
seedream-5-liteImages up to 4K at one price for every size, and edits with up to 14 referencestext, image3
wan-2.7-imageImages and sets of up to 12 related images, up to 4K on Protext, image3, price varies
flux-kontext-proEditing an image with a text instruction while the rest stays puttext, image4
nano-banana-2Images up to 4K with legible text, many references and web searchtext, image4, price varies
nano-banana-2-liteFast, cheap drafts and quick edits at 1Ktext, image4
restore-imageRepairing old, scratched or faded photosimage4
seedream-5-proDetailed photoreal images and edits from up to 10 references, a fee for each after the firsttext, image4, price varies
gpt-image-2Photorealism and exact text in images, such as posters and UI mockupstext, image6
imagen-4-ultraDetailed photorealistic images from a text prompttext6
nano-banana-proHigh-fidelity images up to 4K with accurate text and compositingtext, image9, price varies
qwen-image-layeredSplitting an image into separate transparent layers for editingtext, image11, price varies

Video models 19

Model idBest forTakesCredits
wan-2.2Low-cost short clips, with LoRA supporttext, image1, price varies
seedance-2-miniLower-cost video with a soundtrack from text or images, up to 15 stext, image, video, audio20.5, price varies4.1 per second
veo3.18 s video with native sound from text, first and last frames, or referencestext, image30, price varies
vidu-q4Animating one image into a clip up to 16 s at up to 4K, with soundtext, image38, price varies7.6 per second
vidu-q4-referenceClips up to 16 s at up to 4K that keep characters and voices from up to 15 images and 3 audio clipstext, image, audio38, price varies7.6 per second
wan-3.0Clips up to 30 s at 1080p with sound, from text, frames or image, video and audio referencestext, image, video, audio40, price varies8 per second
kling-3.0-turboFaster Kling clips up to 15 s at 1080p, from text or one imagetext, image45, price varies9 per second
kling-3.0Clips up to 15 s at up to 4K with sound, from text or first and last framestext, image50, price varies10 per second
motion-controlMaking a character from a photo copy the moves in a reference video, billed by that video's lengthtext, image, video50, price varies10 per second
wan-2.6Multi-shot video from text or an image, 5 to 15 s at up to 1080ptext, image, video53, price varies
wan-2.7Video from text or images, continuing a clip, or editing a videotext, image, video, audio60, price varies
wan-3.0-primeFaster clips up to 30 s at 1080p with sound, from text, frames or image, video and audio referencestext, image, video, audio63, price varies12.6 per second
seedance-pro1080p video from a prompt, 2 to 12 stext75, price varies15 per second
seedance-2Video with a soundtrack from text or images, up to 15 s, with a fast tiertext, image, video, audio102.5, price varies20.5 per second
veo3-fastFaster, cheaper 8 s video with native soundtext120
seedance-2.5Clips up to 30 s at 1080p, steered by image, video or audio referencestext, image, video, audio160, price varies32 per second
hunyuanText-to-video with control over frame count, size and stepstext250
veo2Short clips from a prompt or a first-frame imagetext, image250, price varies50 per second
veo38 s video with native sound and dialogue at up to 1080ptext320

Questions

Can one API key call Veo 3.1, Seedance and Nano Banana?

Yes. A key from your profile page calls every model in the table below, veo3.1, the seedance models and the nano-banana models included, and each call bills the same credit balance. There's no plan to buy first and no per-model signup.

Do API credits expire?

No, the balance never expires and never resets. Replicate's prepaid credit, for comparison, is valid for a year from purchase and fal.ai's expires after 365 days.

Do I need an API key to use the MCP server?

Not with a client that supports MCP authorization, which covers Claude, ChatGPT, Claude Code, Codex, Gemini CLI, Cursor and VS Code. The client opens an Azurade page, you sign in and click Allow, and it's connected. A client without OAuth support sends a key from your profile page as Authorization: Bearer sk-...

How do I disconnect a client?

Each connection is a key named after the client, for example "Claude (connected app)", on your profile page. Delete it and the client loses access at once.

What does a generation cost through the API?

The same credits as in the web app. There's no separate developer plan or per-seat fee. A failed generation is refunded to your balance automatically.

Why does create_generation sometimes answer pending?

Most MCP clients give up on a tool call after 60 seconds, so the server returns after about 50 with the job's id. The agent calls get_generation, which waits another 50 seconds. Images usually finish inside the first call; a video can take a few.

Can the agent use an image from my computer?

Yes. create_upload_url gives the agent a link and a curl command; it uploads the file from your machine and passes the returned URL to the model. The link works for 15 minutes and needs no key. A file that is already online needs no upload.

What happens when the agent runs out of credits?

It calls buy_credits and hands you a checkout link for a credit pack. You pay by card on the checkout page, which shows the price, and the credits land on your balance at once. The agent cannot pay by itself.

Can an agent without a key sign itself up?

It can ask for a key, but you approve it. The agent gives you a link and a six-digit code; you sign in, type the code, and the agent receives a key named "Agent" on your account. Delete that key on your profile page to cut it off.