Phosor
Browse documentationLoRA Integration

LoRA Integration Guide

API version: v1.2.4

Learn how to use custom LoRA models to generate videos with unique styles.

What is LoRA?

LoRA (Low-Rank Adaptation) models allow you to customize video generation with specific styles, characters, or visual effects. Each LoRA consists of two files:

  • High noise file: Used during early denoising steps
  • Low noise file: Used during later denoising steps

Step 1: Import Your LoRA

You can import LoRA models either by uploading files or providing URLs.

Option A: Import from URLs

Python
import requests

API_KEY = "your-api-key-here"
API_URL = "https://phosor.ai"

# Import LoRA from URLs
response = requests.post(
    f"{API_URL}/api/v1/storage/lora/import",
    headers={
        "X-API-Key": API_KEY,
        "Content-Type": "application/json"
    },
    json={
        "high_noise_url": "https://example.com/high_noise.safetensors",
        "low_noise_url": "https://example.com/low_noise.safetensors",
        "name": "My Custom Style",
        "default_scale": 1.0
    }
)

result = response.json()
lora_id = result["lora_id"]
print(f"LoRA import started! ID: {lora_id}")

Option B: Upload Files

Python
import requests

API_KEY = "your-api-key-here"
API_URL = "https://phosor.ai"

# Upload LoRA files
with open("high_noise.safetensors", "rb") as high_f, \
     open("low_noise.safetensors", "rb") as low_f:

    response = requests.post(
        f"{API_URL}/api/v1/storage/lora/upload",
        headers={"X-API-Key": API_KEY},
        files={
            "high_noise_file": high_f,
            "low_noise_file": low_f
        },
        data={
            "name": "My Custom Style",
            "default_scale": 1.0
        }
    )

result = response.json()
lora_id = result["lora_id"]
print(f"LoRA upload started! ID: {lora_id}")

Step 2: Wait for LoRA to be Ready

After importing, the LoRA needs processing time. Poll the status endpoint until it's ready.

Python
import time

def wait_for_lora(lora_id, api_key, timeout=300):
    """Wait for LoRA to be ready"""
    start_time = time.time()

    while time.time() - start_time < timeout:
        response = requests.get(
            f"{API_URL}/api/v1/loras/{lora_id}/status",
            headers={"X-API-Key": api_key}
        )

        status = response.json()["status"]
        print(f"Status: {status}")

        if status == "ready":
            print("LoRA is ready!")
            return True
        elif status == "failed":
            print("LoRA processing failed")
            return False

        time.sleep(5)

    print("Timeout waiting for LoRA")
    return False

# Wait for the LoRA to be ready
wait_for_lora(lora_id, API_KEY)

Step 3: Generate Video with LoRA

Once the LoRA is ready, use it in your video generation request.

Python
# Submit video generation with LoRA
response = requests.post(
    f"{API_URL}/api/v1/inference/submit",
    headers={
        "X-API-Key": API_KEY,
        "Content-Type": "application/json"
    },
    json={
        "prompt": "A character walking through a garden",
        "lora_id": lora_id,
        "lora_scale": 0.8,
        "num_frames": 81,
        "frames_per_second": 16
    }
)

request_id = response.json()["request_id"]
print(f"Video generation started! Request ID: {request_id}")

Complete Example

Here's the full workflow in one script:

Python
import requests
import time

API_KEY = "your-api-key-here"
API_URL = "https://phosor.ai"

def import_lora_from_urls(high_url, low_url, name):
    """Import LoRA from URLs"""
    response = requests.post(
        f"{API_URL}/api/v1/storage/lora/import",
        headers={
            "X-API-Key": API_KEY,
            "Content-Type": "application/json"
        },
        json={
            "high_noise_url": high_url,
            "low_noise_url": low_url,
            "name": name,
            "default_scale": 1.0
        }
    )
    return response.json()["lora_id"]

def wait_for_lora(lora_id, timeout=300):
    """Wait for LoRA to be ready"""
    start_time = time.time()

    while time.time() - start_time < timeout:
        response = requests.get(
            f"{API_URL}/api/v1/loras/{lora_id}/status",
            headers={"X-API-Key": API_KEY}
        )

        status = response.json()["status"]

        if status == "ready":
            return True
        elif status == "failed":
            return False

        print(f"Status: {status}, waiting...")
        time.sleep(5)

    return False

def generate_video_with_lora(prompt, lora_id, lora_scale=0.8):
    """Generate video using LoRA"""
    response = requests.post(
        f"{API_URL}/api/v1/inference/submit",
        headers={
            "X-API-Key": API_KEY,
            "Content-Type": "application/json"
        },
        json={
            "prompt": prompt,
            "lora_id": lora_id,
            "lora_scale": lora_scale,
            "num_frames": 81,
            "frames_per_second": 16
        }
    )
    return response.json()["request_id"]

# Full workflow
print("Step 1: Importing LoRA...")
lora_id = import_lora_from_urls(
    "https://example.com/high_noise.safetensors",
    "https://example.com/low_noise.safetensors",
    "My Custom Style"
)
print(f"LoRA ID: {lora_id}")

print("\nStep 2: Waiting for LoRA to be ready...")
if wait_for_lora(lora_id):
    print("\nStep 3: Generating video...")
    request_id = generate_video_with_lora(
        "A character in custom style walking in a park",
        lora_id,
        lora_scale=0.8
    )
    print(f"Video generation started! Request ID: {request_id}")
else:
    print("Failed to prepare LoRA")

LoRA Parameters

ParameterTypeDefaultDescription
lora_idstring-UUID of the LoRA model (required)
lora_scalefloat1.0Strength of LoRA effect (0.0-1.0)

lora_scale tips:

  • 1.0: Full LoRA effect
  • 0.5-0.8: Balanced blend with base model
  • 0.1-0.4: Subtle influence

List Your LoRAs

View all your uploaded LoRA models:

Python
response = requests.get(
    f"{API_URL}/api/v1/loras",
    headers={"X-API-Key": API_KEY}
)

loras = response.json()["loras"]
for lora in loras:
    print(f"- {lora['name']} ({lora['lora_id']}): {lora['status']}")

Next Steps