How to Configure a Custom API Endpoint for a New Model in MuapiClient

Add a new entry with a custom endpoint field to the model catalog in src/lib/models.js, then reference the model's id when calling muapi.generateImage() to automatically route requests to your specified API path.

The Open-Generative-AI repository uses a dynamic model catalog to resolve API endpoints at runtime. When you invoke a generation method, the Muapi client checks the model registry in src/lib/models.js to determine the destination URL. This architecture lets you register new models and their custom endpoints without modifying the core client logic.

Understanding the Endpoint Resolution Logic

The Muapi client builds request URLs dynamically based on the model definition you provide. According to the source code in src/lib/muapi.js (lines 30‑33), the endpoint resolution follows this precedence:

const endpoint = modelInfo?.endpoint || params.model;

Here is the lookup process:

1. Model Lookup: The client searches the exported t2iModels array in src/lib/models.js for an entry matching the model parameter you passed.
2. Endpoint Extraction: If found, it reads the endpoint field from that model's configuration object.
3. Fallback Behavior: If no matching model definition exists, the client uses the raw model string you provided as the endpoint path.

The final request URL concatenates the Muapi base URL (https://api.muapi.ai/api/v1/) with this resolved endpoint value.

Step-by-Step Configuration Guide

Follow these steps to register a new model with a custom API endpoint in the MuapiClient.

Define the Model in models.js

Open src/lib/models.js and locate the exported array beginning with export const t2iModels = [. Append a new object containing at least these required fields:

• id: A unique string identifier used to reference the model in code.
• name: A human-readable label displayed in the Studio UI.
• endpoint: The custom API path segment appended to the base URL.
• inputs: A schema object describing the parameters your endpoint accepts.

Sync the Studio UI (Optional)

The Studio interface reads from packages/studio/src/models.js, which is auto-generated from the source file. After editing src/lib/models.js, rebuild the project or run the generation script (commonly npm run generate-models) to populate the dropdown menus with your new model.

Invoke the Model in Your Code

When calling generation methods like muapi.generateImage(), pass the id of your new model as the model parameter. The client will automatically resolve this to your custom endpoint.

Code Examples

Adding a Custom Model Definition

Add this object to the t2iModels array in src/lib/models.js:

// src/lib/models.js
{
  id: "my-custom-diffusion",
  name: "My Custom Diffusion Model",
  endpoint: "custom-diffusion-v2/generate",
  inputs: {
    prompt: {
      title: "Prompt",
      name: "prompt",
      type: "string",
      description: "Text prompt for image generation",
      examples: ["A cyberpunk cityscape"]
    },
    guidance_scale: {
      title: "Guidance Scale",
      name: "guidance_scale",
      type: "number",
      default: 7.5
    }
  }
}

The endpoint value "custom-diffusion-v2/generate" will be combined with the base URL to create https://api.muapi.ai/api/v1/custom-diffusion-v2/generate.

Calling the Custom Endpoint

Reference the model by its id in your application code:

import { muapi } from '../lib/muapi.js';

async function generateImage() {
  const result = await muapi.generateImage({
    model: 'my-custom-diffusion',  // Matches the id in models.js
    prompt: 'A futuristic skyline at sunset',
    guidance_scale: 8.0
  });
  
  return result.url;
}

Alternatively, you can bypass the model catalog entirely by passing the endpoint string directly:

// Uses the string as the endpoint since no modelInfo exists
const result = await muapi.generateImage({
  model: 'custom-diffusion-v2/generate',
  prompt: 'Abstract art'
});

Key Implementation Details

File	Purpose	Critical Lines
src/lib/muapi.js	Core client handling HTTP requests	Lines 30‑33: const endpoint = modelInfo?.endpoint || params.model;
src/lib/models.js	Master registry of all supported models	Contains export const t2iModels = [...] with objects defining id, name, endpoint
packages/studio/src/models.js	Studio UI model list	Auto-generated mirror of src/lib/models.js

Ensure your endpoint value exactly matches the path exposed by the Muapi service, as the resolution is case-sensitive. The client performs no deep validation of input parameters against your schema, so mismatched fields will result in a 400 error from the server rather than a client-side warning.

Summary

• Register models by adding objects to the t2iModels array in src/lib/models.js.
• Define custom endpoints using the endpoint field in your model configuration.
• Reference by ID when calling methods like muapi.generateImage({ model: 'your-id' }).
• Bypass the catalog by passing the endpoint string directly if you prefer not to register a formal definition.
• Rebuild the UI to see new models in the Studio dropdown by regenerating packages/studio/src/models.js.

Frequently Asked Questions

What file contains the model definitions in the Open-Generative-AI repository?

The master list lives in src/lib/models.js, which exports an array named t2iModels. Each object in this array defines the id, name, endpoint, and input schema for a specific model supported by the MuapiClient.

Can I use a custom endpoint without adding it to models.js?

Yes. If you pass an endpoint string that doesn't match any id in the catalog, the client uses that string directly as the endpoint path due to the fallback logic modelInfo?.endpoint || params.model in src/lib/muapi.js. However, registering the model provides benefits like input validation schemas and Studio UI integration.

How does the MuapiClient construct the final URL for API requests?

The client concatenates the base URL (https://api.muapi.ai/api/v1/) with the resolved endpoint value. The endpoint comes either from the endpoint field of a matched model definition in src/lib/models.js or falls back to the raw model parameter you provided.

Why isn't my new model appearing in the Studio UI dropdown?

The Studio application reads from packages/studio/src/models.js, which is generated from the master src/lib/models.js file. You must rebuild the project or run the model generation script (typically npm run generate-models) to sync the changes, then restart the development server for the UI to reflect the updated catalog.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}