minimax-laravel

A thin Laravel client for the Minimax AI platform. Wraps the four Minimax surfaces we use day-to-day — text-to-image, text-to-speech, chat completions, and voice management (design, list, delete) — behind a single Minimax facade with sensible Laravel defaults (config, env, service container singletons, exception type).

It is deliberately small: each client is a few dozen lines, takes a prompt or text, and returns plain PHP — URLs, bytes, strings, arrays. There is no retry layer, no queue, no SDK-style fluent builder. Drop the result into a job, a Storage disk, an Inertia response, whatever.

What's in the box

All clients raise RayzenAI\Minimax\Exceptions\MinimaxException on HTTP failures or non-zero base_resp.status_code responses.

Install (path repository)

This package is not on Packagist. In the consuming app's composer.json:

{
    "repositories": [
        { "type": "path", "url": "../../rayzenai/minimax-laravel" }
    ],
    "require": {
        "rayzenai/minimax-laravel": "@dev"
    }
}

Then:

composer require rayzenai/minimax-laravel
php artisan vendor:publish --tag=minimax-config

Env

MINIMAX_API_KEY=
MINIMAX_GROUP_ID=
MINIMAX_BASE_URL=https://api.minimax.io/v1

MINIMAX_IMAGE_MODEL=image-01
MINIMAX_IMAGE_ASPECT=9:16
MINIMAX_IMAGE_RESPONSE_FORMAT=url

MINIMAX_TTS_MODEL=speech-2.8-hd
MINIMAX_TTS_VOICE=English_radiant_girl

MINIMAX_TEXT_MODEL=MiniMax-M2.7

# Plan-tier custom voice cap (designed + cloned). Bump for higher tiers.
MINIMAX_VOICE_SLOTS=3

Usage

use RayzenAI\Minimax\Facades\Minimax;
use Illuminate\Support\Facades\Storage;

// 10 portrait images from one prompt.
$urls = Minimax::image()->generate('a cyberpunk samurai at dusk', 10, [
    'aspect_ratio' => '9:16',
]);

// Voiceover bytes — persist however you like.
$mp3 = Minimax::tts()->synthesize('Hello, world.', ['voice' => 'English_radiant_girl']);
Storage::disk('public')->put('hello.mp3', $mp3);

// Chat completion.
$copy = Minimax::text()->complete('Write a 30-second TikTok hook about coffee.');

// Design a new custom voice.
$voice = Minimax::voice()->design(
    prompt: 'A warm, deep British male narrator. Calm and reassuring.',
    previewText: 'Welcome back to the channel.',
);
// $voice = ['voice_id' => '…', 'trial_audio' => <raw mp3 bytes>]

// List + delete.
$voices = Minimax::voice()->list();          // ['system' => […], 'cloning' => […], 'generation' => […]]
Minimax::voice()->delete($voice['voice_id'], 'voice_generation');

Notes & gotchas

• Image URLs expire (~24h) on Minimax's side. Persist bytes if you need long-term storage — the package returns whatever response_format you ask for (URL or data: URI when base64).
• Cloned voices stay invisible until first use — they only appear in the voice_cloning bucket of Minimax::voice()->list() after you've used them in a TTS call once.
• Custom voice cap is plan-tier, not API. Entry plans allow 3. The package doesn't enforce this; minimax.voice_slots is exposed so the consuming app can guard before calling design().
• Voice IDs in TtsClient::VOICES are a curated subset — the live system list from Minimax::voice()->list('system') is canonical.