Foundry API: Black Forest Labs Mismatch

Question

Foundry API: Black Forest Labs Mismatch

Anthony 0

I'm currently using Foundry to host the Black Forest Labs model, FLUX.2-pro.

Under this page, https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/models-sold-directly-by-azure?tabs=global-standard-aoai%2Cglobal-standard&pivots=azure-direct-others,
it states the API should be the same as Black Forest Lab's API, found here: https://docs.bfl.ai/flux_2/flux2_image_editing

For the parameter "input_image", a string type, it states it is:
"Base64 encoded image or URL of image to use as reference. Supports up to 20MB or 20 megapixels. Input resolution: minimum 64x64, recommended up to 2MP, maximum 4MP (e.g., 2048x2048). Dimensions must be multiples of 16."

However, when I invoke the a request to the target URI endpoint for the model, I get an error (in Supabase edge function, which we use for our backend):

"error":{"code":"Invalid input","status":422,"message":"invalid input error","details":[{"type":"Invalid Input","msg":"input_image is a not valid base64 encoded image."}]}}
Please check this guide to understand why this error code might have been returned 
https://docs.microsoft.com/en-us/azure/machine-learning/how-to-troubleshoot-online-endpoints#http-status-codes

I'm a bit confused, since according to Black Forest Labs, using a image URL is acceptable. Is there a way to use an image URL in the request, or is this not possible when using Foundry? Or am I just missing something in our setup?

Thank you very much.

Anthony 0 Reputation points

2026-04-15T05:13:59.9833333+00:00

Also, in our requests to the target uri include webhook_url and webhook_secret parameters which are supported by BFL, per their documentation. However, it seems as though the webhooks are being ignored and the response data (generated images in base64) are being returned synchronously. Are webhooks supported on Foundry for Black Forest labs models? Specifically Flux 2 Pro?
Manas R Mohanty 17,270 Reputation points Moderator

2026-04-18T05:30:54.4533333+00:00
Hi Anthony

Sorry for the inconvenience. It seems you are not able to use Base 64 format of Image URL directly.

You can run these models through the BFL service provider API and through the images/generations and images/edits endpoints.

Azure Foundry endpoint for image editing

https://<resource-name>/openai/deployments/{deployment-id}/images/edits

BFL endpoint

<resource-name>/providers/blackforestlabs/v1/flux-kontext-pro?api-version=preview

On the webhook front, although BFL’s public API supports webhook_url + webhook_secret, the Foundry wrapper for Black Forest Labs models doesn’t currently wire those through. You might always get a synchronous response back from the Foundry endpoint.

Please share your sample code in private message for review

Thank you.

1 answer

Your answer

Anthony 0 Reputation points

2026-04-15T05:13:59.9833333+00:00

Also, in our requests to the target uri include webhook_url and webhook_secret parameters which are supported by BFL, per their documentation. However, it seems as though the webhooks are being ignored and the response data (generated images in base64) are being returned synchronously. Are webhooks supported on Foundry for Black Forest labs models? Specifically Flux 2 Pro?
Manas R Mohanty 17,270 Reputation points Moderator

2026-04-18T05:30:54.4533333+00:00

Hi Anthony

Sorry for the inconvenience. It seems you are not able to use Base 64 format of Image URL directly.

You can run these models through the BFL service provider API and through the images/generations and images/edits endpoints.

Azure Foundry endpoint for image editing

https://<resource-name>/openai/deployments/{deployment-id}/images/edits

BFL endpoint

<resource-name>/providers/blackforestlabs/v1/flux-kontext-pro?api-version=preview

On the webhook front, although BFL’s public API supports webhook_url + webhook_secret, the Foundry wrapper for Black Forest Labs models doesn’t currently wire those through. You might always get a synchronous response back from the Foundry endpoint.

Please share your sample code in private message for review

Thank you.

Answer 1

Foundry exposes FLUX models through a provider-specific BFL API endpoint, but the behavior is constrained by how the model is integrated into Azure. In the provided documentation, input_image is only explicitly described for FLUX.1 Kontext [pro] in the provider-specific API, and it is documented as an image input (with parameters like input_image, aspect_ratio, etc.), not as a URL string.

For FLUX.2 models in Foundry (including FLUX.2-pro):

The BFL provider-specific API is the supported route for fine-grained control.
The documentation calls out that these models accept text and image input and that multi-reference image editing uses base64-encoded images passed alongside the prompt.
The error returned by the Foundry endpoint ("input_image is a not valid base64 encoded image") indicates that the Foundry-hosted FLUX.2-pro endpoint expects input_image to be a valid base64-encoded image payload, not a URL.

Because of this, using a direct image URL for input_image is not supported on the Foundry-hosted FLUX.2-pro endpoint. The integration requires base64-encoded image data in the request body.

To make the request succeed against the Foundry FLUX.2-pro endpoint:

Fetch the image in the backend (for example, in the Supabase edge function).
Convert the binary image data to a base64 string.
Pass that base64 string as the input_image (or reference image field) in the JSON payload to the Foundry BFL endpoint.

If multiple reference images are needed, follow the documented multi-reference pattern for FLUX.2 models (up to eight reference images for FLUX.2-pro), each provided as base64-encoded image data.

References: