About the modelFLUX.1-dev

Hmm… I think this is because the way the service works has changed quite a bit:

The 410 Gone here does not appear to mean that black-forest-labs/FLUX.1-dev itself has been removed.

The important part of the error is:

no longer supported by provider hf-inference

Your URL is explicitly pinning the request to the hf-inference provider:

https://router.huggingface.co/hf-inference/models/black-forest-labs/FLUX.1-dev
                                  ^^^^^^^^^^^^

FLUX.1-dev still exists on the Hub, and at the moment its model page shows fal as a Text-to-Image Inference Provider. Hugging Face’s current Fal provider documentation also uses black-forest-labs/FLUX.1-dev directly in its text-to-image example.

So I would not try to repair this by changing Angular-specific code around the old /hf-inference/ request. I would move to the current Inference Providers client and stop hard-coding hf-inference.

For example, on a server/backend:

import { InferenceClient } from "@huggingface/inference";

const client = new InferenceClient(process.env.HF_TOKEN);

const imageBlob = await client.textToImage({
  model: "black-forest-labs/FLUX.1-dev",
  inputs: "A cinematic photograph of a mountain lake at sunrise",
});

The current JavaScript client uses automatic provider selection by default, so this lets Hugging Face choose an available provider instead of forcing the request through hf-inference. The official Inference Providers documentation currently shows essentially this same FLUX.1-dev JavaScript example.

If you specifically want to pin the currently listed Fal route, you can also make that explicit:

const imageBlob = await client.textToImage({
  provider: "fal-ai",
  model: "black-forest-labs/FLUX.1-dev",
  inputs: "A cinematic photograph of a mountain lake at sunrise",
});

I would prefer the first form unless you have a reason to require a particular provider.

One important difference from the old Serverless Inference API model: Inference Providers are pay-as-you-go beyond the included monthly credits. At the time of writing, the pricing documentation lists $0.10/month of Inference Providers credits for Free accounts (explicitly marked as subject to change), $2/month for PRO, and PAYG after the included credits are exhausted. So changing providers restores a serving route, but it should not be understood as a drop-in replacement for unlimited/free old-style serverless inference.

Also, because you mentioned Angular: if your Hugging Face token is currently stored in the Angular application itself, I would change that separately. This is not the cause of the 410, but tokens embedded in an Angular frontend are visible to the browser user. Both the Hugging Face JavaScript documentation and Angular’s environment documentation recommend keeping secrets server-side and using a proxy/backend.

A safer layout is therefore roughly:

Angular application
        |
        | POST { prompt: ... }
        v
your backend / serverless function
        |
        | HF_TOKEN stays here
        v
@huggingface/inference
        |
        v
Hugging Face Inference Providers
        |
        v
currently available provider

Why this changed / why hf-inference is different now So, for this particular error, my default path would be:

stop pinning /hf-inference/
        ->
use @huggingface/inference
        ->
let provider selection be automatic
        ->
or explicitly use a currently supported provider such as fal-ai
        ->
check the PAYG implications before relying on it in the application

That should separate the actual serving change from the Angular integration, rather than treating the 410 as a problem with FLUX.1-dev itself.