Muapi models

Muapi is one API (

x-api-key

) in front of 500+ models: image, video, audio and speech, lipsync, 3D, LLM text, SEO/data tools, social publishing and utilities. Provider names are hidden; you only see Muapi model names, categories, schemas and costs. Model names are read live, so new models work immediately. Never rely on a remembered model list.

Workflow: discover -> inspect -> estimate -> run -> poll -> download.

Setup

Create a key at https://muapi.ai/dashboard (or a free sandbox key, see below).

Export it and restart your agent from that same shell:

macOS/Linux:
```
export MUAPI_API_KEY=your_key
```

Windows PowerShell:

$env:MUAPI_API_KEY = "your_key"

and

setx MUAPI_API_KEY "your_key"

Keep the key in the environment only. Never write it into project files, commits or chat.

Sandbox key (returns mock outputs instantly, no credits spent), useful for testing a flow:

bash

curl -s -X POST https://api.muapi.ai/api/v1/keys -H "Content-Type: application/json" -d '{"is_test": true}'

Auth is always the

x-api-key: $MUAPI_API_KEY

header. If the

muapi

CLI is installed (

pip install muapi-cli

npm i -g muapi-cli

) prefer it, since it handles upload, polling and

--output-json

. Otherwise use the HTTP calls below (

curl

; on Windows PowerShell use

curl.exe

Rules

User instructions first, then any tool they already configured, then Muapi.
Always inspect before you run. Never guess fields or endpoint paths; schemas change.
Estimate cost before spending credits on anything non-trivial; check balance if no budget was given.
Submit, then poll every 3-10 seconds. Don't hold one long request open.
On
```
failed
```
, show the
```
error
```
field verbatim. Don't silently retry.
Name the model explicitly if the user did; otherwise state which model you picked and why.

1. Discover

bash

curl -s "https://api.muapi.ai/api/v1/models?category=Text%20to%20Image"
curl -s "https://api.muapi.ai/api/v1/models?family=seo"

There is no free-text search param: fetch a category and filter client-side by name/description, and try a few keyword variants. Categories:

Text to Image

Image to Image

Text to Video

Image to Video

Video to Video

Audio to Video

(lipsync),

Text to Audio

Image to 3D

Text to 3D

Text to Text

(LLMs and data tools),

Lora Support

Training

. CLI:

muapi discover "cinematic image to video" --output-json

2. Inspect

bash

curl -s "https://api.muapi.ai/api/v1/models/{name}"

Returns the input schema (required fields, allowed values, defaults), pricing and the real

endpoint

to POST to. CLI:

muapi inspect {name} --output-json

3. Estimate cost

bash

curl -s -X POST "https://api.muapi.ai/api/v1/models/{name}/estimate-cost" \
  -H "Content-Type: application/json" -d @payload.json

Same body you would send to run it. Balance:

GET /api/v1/account/balance

(CLI:

muapi account balance

4. Upload input files

Models that take an image, video or audio need a hosted URL. Upload local files first:

bash

curl -s -X POST https://api.muapi.ai/api/v1/upload_file \
  -H "x-api-key: $MUAPI_API_KEY" -F "file=@./input.png"

Limits: images 10MB, videos 50MB. CLI:

muapi upload file ./input.png --output-json

5. Run

POST to the

endpoint

from step 2 (never a path you assumed):

bash

curl -s -X POST "https://api.muapi.ai{endpoint}" \
  -H "x-api-key: $MUAPI_API_KEY" -H "Content-Type: application/json" -d @payload.json
# -> {"request_id": "...", "status": "processing"}

CLI:

muapi run {name} --input payload.json --no-wait --output-json

6. Poll

bash

curl -s "https://api.muapi.ai/api/v1/predictions/{request_id}/result" -H "x-api-key: $MUAPI_API_KEY"

status

processing

completed

failed

. On

completed

outputs

is an array of URLs. CLI:

muapi runs get {request_id} --wait --output-json

7. Save results

Result URLs are kept for 30 days, then expire. As soon as a task completes, download every output to the location the user asked for (for example

./public/hero.png

) and tell them the path:

bash

curl -L -o ./public/hero.png "{output_url}"

If the user wants it in their app, write the integration code instead of just a file.

Writing good requests

Ask for or infer: the output destination, a specific model if they have a preference, and key parameters (aspect ratio, resolution, duration). Examples:

"Make a 1:1 product photo of a white ceramic mug with Muapi, save to ./public/mug.png"
"Generate a 30 second lo-fi background track with Muapi"
"Turn README.md into a spoken intro with a Muapi speech model"

Troubleshooting

Problem	Fix
`MUAPI_API_KEY` missing	Export it and restart the agent from the same terminal
401	Check the key at https://muapi.ai/dashboard; header must be `x-api-key`
Insufficient credits	Top up in the dashboard, then rerun
`failed` task	Show `error` , adjust parameters from the schema, retry once
Expired result link (older than 30 days)	Re-run, and save outputs locally next time
`curl` rejects args on Windows	Use `curl.exe`

Other interfaces

OpenAPI: https://api.muapi.ai/openapi.json · MCP: https://muapi.ai/.well-known/mcp.json · Human docs: https://muapi.ai/agent-skills

muapi-models

NPX Install

Tags

SKILL.md Content

Muapi models

Setup

Rules

1. Discover

2. Inspect

3. Estimate cost

4. Upload input files

5. Run

6. Poll

7. Save results

Writing good requests

Troubleshooting

Other interfaces