Agent M1

warning

Please test this agent on your own use-cases. It was tested and measured by only the official benchmarks in the GitHub repository FlyMyAI/bench_M1.

Overview

FlyMy.AI Media Agent 1 (M1) is a unified multimodal media agent that provides access to dozens of high-performance media models through a single API endpoint. The agent combines open-source innovations with proprietary models to deliver instant media generation capabilities, including real-time image generation, face-preserving editing, and video creation. M1 aims to simplify the integration of modern AI models for developers.

Key Capabilities

For detailed benchmark results and comparisons with other models, please visit our benchmark results page.

Image Generation

Agent M1 performance in image generation tasks:

• Single object generation: 99%
• Multiple object composition: 97%
• Object counting: 70%
• Color understanding: 86%
• Positional awareness: 51%
• Color attribute handling: 58%

Face Identity Preservation

Agent M1 performance in face transformations:

• Overall identity preservation: 91.7%
• Emotion transformations: 97.7%
• Age transformation: 91.5%
• Hair style modifications: 89.9%
• Accessory additions: 93%

World Knowledge Understanding

Agent M1 performance across domains:

• Cultural knowledge: 79.1%
• Temporal understanding: 92.6%
• Spatial reasoning: 87.6%
• Biological concepts: 83.8%
• Physical principles: 91%
• Chemical knowledge: 84.1%

Usage

Basic Usage

import requests
import json

# Initialize the API client
API_URL = "https://api.flymyai.com/v1/agent-m1"
headers = {
    "Authorization": f"Bearer {YOUR_API_KEY}",
    "Content-Type": "application/json"
}

# Generate an image
def generate_image(prompt):
    payload = {
        "prompt": prompt,
        "num_images": 1
    }
    response = requests.post(
        f"{API_URL}/generate",
        headers=headers,
        json=payload
    )
    return response.json()

# Transform a face
def transform_face(image_path, transformation):
    with open(image_path, 'rb') as image_file:
        files = {'image': image_file}
        payload = {
            "transformation": transformation
        }
        response = requests.post(
            f"{API_URL}/transform",
            headers={"Authorization": f"Bearer {YOUR_API_KEY}"},
            files=files,
            data=payload
        )
    return response.json()

# Get world knowledge response
def ask_question(question):
    payload = {
        "question": question
    }
    response = requests.post(
        f"{API_URL}/ask",
        headers=headers,
        json=payload
    )
    return response.json()

# Example usage
image_response = generate_image("A serene landscape with mountains and a lake")
face_response = transform_face("path_to_image.jpg", "Add a smile")
knowledge_response = ask_question("What causes the seasons to change?")

Advanced Usage

# Complex image generation with multiple objects
advanced_payload = {
    "prompt": "A red car parked next to a blue house with a white picket fence",
    "num_images": 1,
    "style": "photorealistic",
    "negative_prompt": "blurry, distorted",
    "parameters": {
        "guidance_scale": 7.5,
        "num_inference_steps": 50
    }
}
response = requests.post(
    f"{API_URL}/generate",
    headers=headers,
    json=advanced_payload
)

# Face transformation with specific attributes
face_payload = {
    "transformation": "Make the person look 20 years older",
    "preserve_identity": True,
    "style": "natural",
    "parameters": {
        "strength": 0.8,
        "preserve_expression": True
    }
}
with open("path_to_image.jpg", 'rb') as image_file:
    files = {'image': image_file}
    response = requests.post(
        f"{API_URL}/transform",
        headers={"Authorization": f"Bearer {YOUR_API_KEY}"},
        files=files,
        data=face_payload
    )

Best Practices

1. Prompt Engineering

   • Be specific and detailed in your prompts
   • Use clear descriptions of objects and their relationships
   • Specify colors and positions when important

2. Face Transformations

   • Provide clear transformation instructions
   • Use natural language descriptions
   • Consider using the preserve_identity parameter for critical applications

3. World Knowledge Queries

   • Frame questions clearly and specifically
   • Provide context when necessary
   • Use follow-up questions for complex topics

Limitations

• Performance may vary based on specific use cases
• Complex scenes with many objects may require multiple attempts
• Face transformations work best with clear, front-facing images
• World knowledge responses are based on training data cutoff

Python SDK Usage

Using Synchronous Client

from flymyai import m1_client

client = m1_client(apikey="fly-secret-key")
result = client.generate("An Iron Man")
print(result.data.text, result.data.file_url)

FlymyAI M1 client also stores request history for later generation context:

from flymyai import m1_client

client = m1_client(apikey="fly-secret-key")

result = client.generate("An Iron Man")
print(result.data.text, result.data.file_url)

result = client.generate("Add him Captain America's shield")
print(result.data.text, result.data.file_url)

Passing image

from pathlib import Path
from flymyai import m1_client

client = m1_client(apikey="fly-secret-key")
result = client.generate("An Iron Man", image=Path("./image.png"))
print(result.data.text, result.data.file_url)

Using Asynchronous Client

import asyncio
from flymyai import async_m1_client


async def main():
    client = async_m1_client(apikey="fly-secret-key")
    result = await client.generate("An Iron Man")
    print(result.data.text, result.data.file_url)


asyncio.run(main())

Passing image

import asyncio
from pathlib import Path
from flymyai import async_m1_client


async def main():
    client = async_m1_client(apikey="fly-secret-key")
    result = await client.generate("An Iron Man", image=Path("./image.png"))
    print(result.data.text, result.data.file_url)


asyncio.run(main())

Support

For technical support or to report issues, please visit our GitHub repository or contact our support team.