Inference Providers

Inference Providers are a simple, hosted way to run models via API without deploying infrastructure. You can route requests automatically to an available provider or choose a specific provider for a model.

In this unit you’ll learn the basics, make a first call, and then build a small app. Keep an eye on the Tips — they highlight common gotchas and best practices.

Key ideas

Quickstart (Python)

from huggingface_hub import InferenceClient

# Create a client. By default, it will auto-pick a provider.
client = InferenceClient()

# Text generation
response = client.text_generation(
    "Write a short poem about the ocean.",
    model="gpt2",  # replace with a model you want to try
)
print(response)

# Image generation (example)
# img_bytes = client.image_generation(
#     prompt="a watercolor painting of a lighthouse at dawn",
#     model="stabilityai/stable-diffusion-2",
# )
# with open("output.png", "wb") as f:
#     f.write(img_bytes)

To prefer a specific provider, pass provider="..." to your call (or when constructing the client). When in doubt, keep provider="auto" to use the first available provider based on your preferences.

Authentication

InferenceClient reads your token from the HF_TOKEN environment variable or you can pass token explicitly:

from huggingface_hub import InferenceClient

client = InferenceClient(token="hf_xxx")
print(client.text_generation("Hello!", model="gpt2"))

cURL example

You can also call the Inference API directly with cURL:

curl -X POST \
  -H "Authorization: Bearer $HF_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"inputs": "Write a haiku about the sea"}' \
  https://api-inference.huggingface.co/models/gpt2

Your first API call (Text-to-Image)

Before coding, pick a model on the Hub with Inference Providers enabled and try the right-side widget. Then replicate the request in code.

Python:

import os
from huggingface_hub import InferenceClient

client = InferenceClient(
    provider="auto",
    api_key=os.environ["HF_TOKEN"],
)

# Returns a PIL.Image
image = client.text_to_image(
    "Astronaut riding a horse",
    model="black-forest-labs/FLUX.1-schnell",
)
image.save("image.png")

TypeScript:

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

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

const image = await client.textToImage({
  provider: "auto",
  model: "black-forest-labs/FLUX.1-schnell",
  inputs: "Astronaut riding a horse",
  parameters: { num_inference_steps: 5 },
});
// image is a Blob

Streaming

For long generations, stream tokens as they arrive:

from huggingface_hub import InferenceClient

client = InferenceClient()
for chunk in client.text_generation(
    "Explain transformers in 2 sentences.",
    model="gpt2",
    stream=True,
):
    print(chunk, end="", flush=True)

Provider selection

provider="auto" is the default and tries providers in your preference order with failover. You can also pin a specific provider.

Python:

from huggingface_hub import InferenceClient

client = InferenceClient()

# Auto selection (default)
img_auto = client.text_to_image(
    "Astronaut riding a horse",
    model="black-forest-labs/FLUX.1-schnell",
    provider="auto",
)

# Explicit provider
img_fal = client.text_to_image(
    "Astronaut riding a horse",
    model="black-forest-labs/FLUX.1-schnell",
    provider="fal-ai",
)

TypeScript:

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

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

const imageAuto = await client.textToImage({
  model: "black-forest-labs/FLUX.1-schnell",
  inputs: "Astronaut riding a horse",
  provider: "auto",
  parameters: { num_inference_steps: 5 },
});

const imageFal = await client.textToImage({
  model: "black-forest-labs/FLUX.1-schnell",
  inputs: "Astronaut riding a horse",
  provider: "fal-ai",
  parameters: { num_inference_steps: 5 },
});

Error handling and timeouts

You’re ready to build an app with providers—let’s do that next.

Up next: build your first AI app using Providers (transcribe + summarize).

< > Update on GitHub