How-To Interact With MirageLSD Video-to-Video Model On Crusoe Intelligence Foundry

Last Updated: Dec 30, 2025

Introduction

Crusoe Managed Inference provides high-performance access to specialized AI models like Decart MirageLSD. This article demonstrates how to transform video content using a text prompt via a managed batch queue. By leveraging our high-throughput infrastructure, users can achieve near-real-time results using standard curl commands as well as Python libraries.

Prerequisites

• Access to Crusoe Intelligence Foundry 

• Terminal access with curl installed OR Python 3.x installed with requests library & basic familiarity with Python scripts

• Input video file

Step-by-Step Instructions

Generate Inference Key

The v2v video transformation process follows a logical lifecycle as explained below. Before that, generate an API Key in the Crusoe Foundry UI.  

1. Click on "Get API Key" on the right top corner

   

2. Enter name, select Crusoe project name and/or set expiration date for the key. Click on Create and Copy the key as its used in the following steps.

   • ⚠️ Important: Expiration date must be set after the current time. By default, keys are valid until 12:00 AM UTC on the chosen date.

   • If expiration is invalid, you will see an error as follows

     Error
     Could not create Inference API Key: undefined

    

   

Interacting with Video to Video Decart MirageLSD Model using Curl 

1. Upload the Input Video 
   
   Use the Files API to upload your source video to Crusoe Cloud storage. You must specify the purpose as "video" to ensure correct processing. 

   Tip: To try the feature out, you could download any one sample videos from here. 

   $ curl -X POST https://ai-api-eu-iceland1-a.crusoecloud.com/v1/files \
     -H 'Authorization: Bearer <API Key>' \
     -H 'Crusoe-Project-Id: <Project ID>' \
     -F "purpose=video" \
     -F "file=@/path/to/your/video.mp4"

   Note: The name of the video must be "video.mp4"

   Example Output

   {"id":"3326b244-c5ea-4b3d-b298-0022f107913f","type":"video_input","bytes":3084764,"created_at":1766044079,"filename":"video.mp4","purpose":"video","status":"uploaded"}

2. Submit the Request to the Queue 
   Send a POST request to the Decart MirageLSD endpoint, referencing your file_id from Step 1 and providing a text prompt for the desired modification.

   $ curl -X POST https://api-video.crusoe.ai/v1/queue/decart/miragelsd-1-batch/enhanced \
     -H 'Authorization: Bearer <API Key>' \
     -H 'Content-Type: application/json' \
     -d '{
       "file_id": "<file_id>",
       "prompt": "Modify the video to show a black horse" <---- update this prompt according to your desire
     }'

   Example Output

   {"request_id":"ca01c37a5b8884e0ac8626c1a1a0368b","status_url":"https://api-video.crusoe.ai/v1/queue/decart/miragelsd-1-batch/requests/ca01c37a5b8884e0ac8626c1a1a0368b","cancel_url":"https://api-video.crusoe.ai/v1/queue/decart/miragelsd-1-batch/requests/ca01c37a5b8884e0ac8626c1a1a0368b/cancel"}% 

3. Monitor Job Status 
   The previous step returns a request_id. Use this to poll the status URL until the job is complete (status 200).

   $ curl -X GET https://api-video.crusoe.ai/v1/queue/decart/miragelsd-1-batch/requests/<Request ID> \
     -H 'Authorization: Bearer <API Key'

   Example Output

   When successful, the response will include a result_file_id as shown:

   {"request_id":"ca01c37a5b8884e0ac8626c1a1a0368b","result_file_id":"b9eedc81-e955-4bc0-be2a-86aea7717447","status":200,"last_updated":1766044329142} 

4. Download the Output Video 
   Once the status indicates completion, use the Files API to download the resulting video to your local machine.

   $ curl -X GET https://api-video.crusoe.ai/v1/files/<result_file_id> \
     -H 'Authorization: Bearer <API Key>' \
     -o /path/to/download/output.mp4

   Example Output

     % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                    Dload  Upload   Total   Spent    Left  Speed
   100 2610k  100 2610k    0     0  1042k      0  0:00:02  0:00:02 --:--:-- 1042k

Interacting with Video to Video Decart MirageLSD Model using Crusoe’s MirageLSD Inference API

1. Upload the Input Video

• Save your video locally.

• Use the following Python script to upload the video. Replace <INPUT_VIDEO_PATH>, <YOUR_API_KEY> and <YOUR_PROJECT_ID> with the correct values

import requests

INPUT_VIDEO_PATH = "<INPUT_VIDEO_PATH>"
AUTH_TOKEN = "<YOUR_API_KEY>"
PROJECT_ID = "<YOUR_PROJECT_ID>"

with open(INPUT_VIDEO_PATH, "rb") as f:
    upload_response = requests.post(
        "https://ai-api-eu-iceland1-a.crusoecloud.com/v1/files",
        headers={
            "Authorization": f"Bearer {AUTH_TOKEN}",
            "Crusoe-Project-Id": PROJECT_ID,
        },
        files={"file": f},
        data={"purpose": "video"},
    )
    upload_response.raise_for_status()

upload_data = upload_response.json()
file_id = upload_data["id"]

print(f"File uploaded successfully. id: {file_id}")

Example Output:

File uploaded successfully. id: dd573be1-1ed1-4c9f-b9c1-de352901df30

Notes / Tips:

• The returned file_id is required for the batch inference request.

• Successful upload confirmation: http_status 200.

Common errors:

• 401 Unauthorized → invalid API key. Check if your key is valid and has not expired.

• Invalid file path → Python FileNotFoundError. Confirm the INPUT_VIDEO_PATH is correct and exists.

2: Submit the MirageLSD Batch Inference Request

• Enqueue a job using the uploaded file_id. Replace <YOUR_API_KEY>, <YOUR_PROJECT_ID> and <UPLOADED_FILE_ID> with the correct values. <UPLOADED_FILE_ID>  is the file_id we received in Step 2. Replace <YOUR_PROMPT_DESCRIBING_THE_DESIRED_VIDEO> with the prompt describing how you want the output video to look

import requests

AUTH_TOKEN = "<YOUR_API_KEY>"
PROJECT_ID = "<YOUR_PROJECT_ID>"
file_id = "<UPLOADED_FILE_ID>"

inference_response = requests.post(
    "https://ai-api-eu-iceland1-a.crusoecloud.com/v1/queue/decart/miragelsd-1-batch/enhanced",
    headers={
        "Authorization": f"Bearer {AUTH_TOKEN}",
        "Crusoe-Project-Id": PROJECT_ID,
        "Content-Type": "application/json",
    },
    json={
        "file_id": file_id,
        "prompt": "<YOUR_PROMPT_DESCRIBING_THE_DESIRED_VIDEO>",
    },
)
inference_response.raise_for_status()

inference_data = inference_response.json()
request_id = inference_data["request_id"]
status_url = inference_data["status_url"]

if not request_id or not status_url:
   raise ValueError("Could not find 'request_id' or 'status_url' in enqueue response.")

print(f"Request enqueued. request_id: {request_id}")
print(f"Status Url: {status_url}")

Example Output:

Request enqueued. request_id: c4b7ff33ab1f5b7ed767a434285f2c25
Status Url: https://ai-api-eu-iceland1-a.crusoecloud.com/v1/queue/decart/miragelsd-1-batch/requests/c4b7ff33ab1f5b7ed767a434285f2c25

Notes / Tips:

• request_id identifies your job.

• status_url is used to poll for job completion.

Common errors:

• 400 Bad Request → invalid file_id. Confirm that file_id is correct

• 401 Unauthorized → invalid API key. Check if your key is valid and has not expired.

3: Poll for Job Completion

• Use the following Python script to poll for output status.  Replace <YOUR_API_KEY>, <YOUR_PROJECT_ID> with the correct values. Replace <STATUS_URL> with the value we received in Step 3.

import time
import requests

is_complete = False
result_file_id = None

AUTH_TOKEN = "<YOUR_API_KEY>"
PROJECT_ID = "<YOUR_PROJECT_ID>"
status_url = "<STATUS_URL>"

while not is_complete:
    status_response = requests.get(
        status_url,
        headers={
            "Authorization": f"Bearer {AUTH_TOKEN}",
            "Crusoe-Project-Id": PROJECT_ID,
        },
    )
    status_response.raise_for_status()
    status_data = status_response.json()
    status = status_data["status"]

    if status == 200:
        is_complete = True
        result_file_id = status_data["result_file_id"]
    else:
        time.sleep(2)

print(f"Request complete. result_file_id: {result_file_id}")

• When complete, you’ll receive a result_file_id that points to the transformed output video.

Example Output:

Request complete. result_file_id: 5426b02e-9897-42c0-a4ef-09c2c4cc07f1

Notes / Tips:

• Polling every 2 seconds is sufficient for near-real-time batch jobs.

• Use the result_file_id to download the processed video.

Common errors:

• 400 Bad Request → incorrect status_url. Verify that it is correct.

4: Download the Output Video

• Use the following Python script to download the output video file.  Replace <YOUR_API_KEY>, <YOUR_PROJECT_ID> with the correct values. Replace <OUTPUT_VIDEO_PATH> with the path where you want to save the file - example output/video_output.mp4. Replace <RESULT_FILE_ID> with the value we received in Step 4.

import requests

AUTH_TOKEN = "<YOUR_API_KEY>"
PROJECT_ID = "<YOUR_PROJECT_ID>"
OUTPUT_VIDEO_PATH = "<OUTPUT_VIDEO_PATH>"
result_file_id = "<RESULT_FILE_ID>"

download_response = requests.get(
    f"https://ai-api-eu-iceland1-a.crusoecloud.com/v1/files/{result_file_id}",
    headers={
        "Authorization": f"Bearer {AUTH_TOKEN}",
        "Crusoe-Project-Id": PROJECT_ID,
    },
)
download_response.raise_for_status()

# Ensure output directory exists
import os
os.makedirs(os.path.dirname(OUTPUT_VIDEO_PATH), exist_ok=True)

with open(OUTPUT_VIDEO_PATH, "wb") as f:
    f.write(download_response.content)

print(f"Output video saved at {OUTPUT_VIDEO_PATH}")

Example Output: 

File downloaded successfully to: output/video_output.mp4

Notes / Tips:

• Make sure the directory in OUTPUT_VIDEO_PATH exists.

Common errors:

• 404 Not Found → invalid result_file_id. Confirm if the result file id is correct

• 401 Unauthorized → invalid API key. Check if your key is valid and has not expired.

• Invalid output path -> Python NotADirectoryError. Make sure the directory where you want to save the output video exists.

5: Combine the entire script

• You can combine Steps 2-5 in a single script. Example is given on the docs here.

Additional Resources

• Decart MirageLSD in Crusoe Foundry UI
• Crusoe v2v Managed Inference models
• Python Requests Library