Flux2-pro deployed in Foundry shows timeout error

Question

Hi,

I'm trying to edit pictures, removing the background. The same code worked last week but now it shows the error {'error': {'code': 'Timeout', 'message': 'The operation was timeout.'}}, status: 408, with images larger than 170 Kb, but last week the same code worked.

Any idea what's happening?

import base64
import requests
from azure.identity import AzureCliCredential
FLUX2_PRO_ENDPOINT = "https://<resource-name>.services.ai.azure.com/providers/blackforestlabs/v1/flux-2-pro?api-version=preview"
image_path = "<image-path.jpg"

credential = AzureCliCredential()
def get_token():
    return credential.get_token("https://cognitiveservices.azure.com/.default")

def get_headers():
    token = get_token()
    return { "Authorization": f"Bearer {token.token}" }

def image_to_base64(image_path):
    """Convertir imagen a base64"""
    with open(image_path, "rb") as f:
        return base64.b64encode(f.read()).decode("utf-8")


PROMPT_WHITE_BG = "remove background and set the background pure white, do not modify the main subject in any way, keep the same position, scale and pose, keep all details of the subject, only change the background to pure white #FFFFFF, keep natural contact shadows where the subject meets the ground, maintain identical camera angle, framing and perspective."

data = {
        "prompt": PROMPT_WHITE_BG,
        "input_image": image_to_base64(image_path),  # Base64 encoded image
        "width": 1024,
        "height": 800,
        "n": 1,
        "model": "FLUX.2-pro"
}

response = requests.post(FLUX2_PRO_ENDPOINT, headers=get_headers(), json=data)
print(response.json())

Answer 1

Hi,

Following a suggestion from a member of the Microsoft support team, I tested the process using the model deployed in the US Data Zone instead of Global Standard, and without resizing the images, all of them were processed successfully with no timeout errors.Regards,

Angel

Answer 2

hi,

Thank you for the detailed answer, awesome!!

I found a workaround: resizing the images to 1792 px or limiting them to 2 MP works for almost all images, and for the remaining ones that still hit a timeout error, resizing them to 1024 px works.

Thanks!

Answer 3

Hello Angel Sevillano,

Welcome to Microsoft Q&A and Thank you for sharing the details and code snippet.

Since the same code was working last week and now returns a 408 Timeout error only for images larger than ~170 KB, this likely points to a service-side processing or latency issue rather than a problem in your implementation.

Below are the most relevant things to check and try:

Image Size and Base64 Payload

When sending images as Base64:

• Base64 increases the payload size by ~33%.

A 170 KB image becomes ~225 KB in the request body.

Larger images significantly increase preprocessing time for diffusion-based models like FLUX.2-pro.

If larger images now time out, this could indicate:

Increased backend latency

Stricter request processing limits

Higher regional load

Recommendation:

Resize images before encoding (e.g., max 1024x1024).

Compress images (quality=85, optimize=True) before converting to Base64.

Test with smaller output resolution (e.g., 768x768) to confirm if this is compute-related.

If smaller images work consistently, this confirms a processing timeout rather than a code issue.

Increase Client Request Timeout

Your current requests.post() call does not specify a timeout. While the service may still enforce limits, increasing the client timeout ensures your client isn’t closing early.

Example:

response = requests.post(
    FLUX2_PRO_ENDPOINT,
    headers=get_headers(),
    json=data,
    timeout=120  # 2 minutes
)

You can test with 120–180 seconds.

Deployment Request Timeout Setting

If this model is deployed through Foundry with configurable deployment settings, verify whether a request timeout is set.

You can increase:

request_settings:
  request_timeout_ms: 300000

(5 minutes)

If the timeout is happening at the deployment layer, this may resolve it.

Service Health Check

Since this worked last week:

Check Azure Service Health for any incidents affecting:

Azure AI Foundry

  BlackForestLabs provider
  
     Your specific region
     

Latency increases due to regional capacity changes can cause previously working workloads to start timing out.

Quota or Throttling

Ensure your subscription has not:

Reached quota limits

Hit throttling thresholds

Experienced reduced regional capacity

Even without explicit 429 errors, backend capacity pressure can manifest as timeouts.

Monitor Performance

Use:

• Azure Monitor
• Diagnostic logs
• Correlation ID from failed requests

Likely Root Cause

Given that:

The same code worked previously

Smaller images still work

Only larger payloads fail

This strongly suggests either:

Increased model latency under load

Reduced processing timeout window

Regional capacity change

Backend behavior change in preview API

Since you are using a preview API version, changes in limits or backend behavior can occur without explicit announcements.

Suggested Diagnostic Test

Try Same image, Resize to 512x512, Keep same prompt.

If it succeeds → confirms compute timeout rather than payload rejection.

If resizing and timeout adjustments do not resolve the issue, this may require escalation to check backend latency or capacity for FLUX.2-pro in your region.

Please refer this

• Deploy models, flows, and web apps with Azure AI Foundry
• Steps for troubleshooting your deployments in Azure AI Foundry

I Hope this helps. Do let me know if you have any further queries.

Thank you!