The BFL Finetuning API enables you to customize FLUX Pro and FLUX Ultra using 1-20 images of your own visual content, and optionally, text descriptions.

​
Regional Requirements for Finetuning

Important: Finetuning operations must use region-specific endpoints and cannot use the global or multi-cluster regional endpoints.

​
Finetuning Endpoints

EU Region: api.eu1.bfl.ai
  • Use this endpoint for all finetuning operations in the EU region
  • Required for GDPR compliance when training with EU data
US Region: api.us1.bfl.ai
  • Use this endpoint for all finetuning operations in the US region

​
Regional Consistency

You can only run inference on finetuned models in the same region where they were trained. When you submit a finetuning task using https://api.us1.bfl.ai/v1/finetune, you must use https://api.us1.bfl.ai/v1/flux-pro-1.1-ultra-finetuned for inference. Cross-region inference will not find your finetuned model.

​
Getting Started: Step-by-Step Guide

1

Prepare Your Images

Create a local folder containing your training images:
  • Supported formats: JPG, JPEG, PNG, and WebP
  • Recommended: More than 5 images
High-quality datasets with clear, articulated subjects/objects/styles significantly improve training results. Higher resolution source images help but are capped at 1MP.
2

Add Text Descriptions (Optional)

Create text files with descriptions for your images:
  • Text files should share the same name as their corresponding images
  • Example: If your image is sample.jpg, create sample.txt
3

Package Your Data

Compress your folder into a ZIP file containing all images and optional text descriptions.
4

Configure Training Parameters

Select appropriate hyperparameters based on your use case. See the Training Parameters section below for detailed configuration options.
5

Submit Training Task

Use the provided Python functions to submit your finetuning task to the BFL API.
6

Monitor Progress

Check the status of your training job using the progress monitoring functions.
7

Run Inference

Once training is complete, use your custom model through the available finetuned endpoints.

​
Model Training Parameters

​
Required Parameters

​
mode
string
required
Determines the finetuning approach based on your concept.Options: "character", "product", "style", "general"
In “general” mode, the entire image is captioned when captioning is True without specific focus areas. No subject specific improvements will be made.
​
finetune_comment
string
required
Descriptive note to identify your fine-tune since names are UUIDs. Will be displayed in finetune_details.

​
Optional Parameters

​
iterations
integer
default:"300"
Minimum: 100Defines training duration. For fast exploration, 100-150 iterations can be enough. For more complex concepts, larger datasets, or extreme precision, more iterations than the default can help.
​
learning_rate
float
default:"0.00001 (full) / 0.0001 (lora)"
Default: 0.00001 if finetune_type is “full”, 0.0001 if finetune_type is “lora”
Lower values can improve the result but might need more iterations to learn a concept. Higher values can allow you to train for less iterations at a potential loss in quality.
For finetune_type “lora”, values 10 times larger than for “full” are recommended.
​
priority
string
default:"quality"
Options: "speed", "quality", "high_res_only"The speed priority will improve speed per training step.
​
captioning
boolean
default:"true"
Enables/disables automatic image captioning.
​
trigger_word
string
default:"TOK"
Unique word/phrase that will be used in the captions to reference the newly introduced concepts.
​
lora_rank
integer
default:"32"
Choose between 32 and 16. A lora_rank of 16 can increase training efficiency and decrease loading times.
​
finetune_type
string
default:"full"
Choose between “full” for a full finetuning + post hoc extraction of the trained weights into a LoRA or “lora” for a raw LoRA training.

​
Inference Endpoints

Available endpoints for your finetuned model:
  1. /flux-pro-1.1-ultra-finetuned
  2. /flux-pro-finetuned
  3. /flux-pro-1.0-depth-finetuned
  4. /flux-pro-1.0-canny-finetuned
  5. /flux-pro-1.0-fill-finetuned

​
Regional Inference Requirements

Regional Consistency Required: You can only inference the finetunes in the region that you trained them in.Querying the global endpoint https://api.bfl.ai/v1/flux-pro-1.1-ultra-finetuned will not find your region-specific finetune.

​
Additional Inference Parameters

The endpoints have additionally all input parameters that their non-finetuned sibling endpoints have.
​
finetune_id
string
required
References your specific model.
Find the finetune_id either in my_finetunes or in the return dict of your /finetune POST.
​
finetune_strength
float
default:"1.0"
Range: 0-2Controls finetune influence. Increase this value if your target concept isn’t showing up strongly enough. The optimal setting depends on your finetune and prompt.

​
Implementation Guide

​
Setup and Dependencies

1

Install Dependencies

pip install requests fire
2

Set API Key

export BFL_API_KEY="your_api_key_here"

​
Implementation Guide

​
Example Python Implementation


Assuming you have prepared your images in a `finetuning.zip` file:
# submit finetuning task
$ python bfl_finetune.py request_finetuning finetuning.zip myfirstfinetune
id:            <finetune_id>

# query status
$ python bfl_finetune.py finetune_progress <finetune_id>
id:       <finetune_id>
status:   Pending
result:   null
progress: null

# once status shows Ready, run inference (defaults to flux-pro-1.1-ultra-finetuned)
$ python bfl_finetune.py finetune_inference <finetune_id> --prompt="image of a TOK"
finetune_id: <inference_id>

# retrieve inference result
$ python bfl_finetune.py get_inference <inference_id>
id:       <inference_id>
status:   Ready
result:   {"sample": <result_url>, "prompt": "image of a TOK"}
progress: null
"""

import os
import base64
import requests


def request_finetuning(
    zip_path,
    finetune_comment,
    trigger_word="TOK",
    mode="general",
    api_key=None,
    iterations=300,
    learning_rate=0.00001,
    captioning=True,
    priority="quality",
    finetune_type="full",
    lora_rank=32,
):
    """
    Request a finetuning using the provided ZIP file.

    Args:
        zip_path (str): Path to the ZIP file containing training data
        finetune_comment (str): Comment for the finetune_details
        trigger_word (str): Trigger word for the model
        mode (str): Mode for caption generation
        api_key (str): API key for authentication
        iterations (int): Number of training iterations
        learning_rate (float): Learning rate for optimization
        captioning (bool): Enable/disable auto-captioning
        priority (str): Training quality setting
        lora_rank (int): Lora rank
        finetune_type (str): "full" or "lora"

    Returns:
        dict: API response

    Raises:
        FileNotFoundError: If ZIP file is missing
        requests.exceptions.RequestException: If API request fails
    """
    if api_key is None:
        if "BFL_API_KEY" not in os.environ:
            raise ValueError(
                "Provide your API key via --api_key or an environment variable BFL_API_KEY"
            )
        api_key = os.environ["BFL_API_KEY"]

    if not os.path.exists(zip_path):
        raise FileNotFoundError(f"ZIP file not found at {zip_path}")

    assert mode in ["character", "product", "style", "general"]

    with open(zip_path, "rb") as file:
        encoded_zip = base64.b64encode(file.read()).decode("utf-8")

    url = "https://api.us1.bfl.ai/v1/finetune"
    headers = {
        "Content-Type": "application/json",
        "X-Key": api_key,
    }
    payload = {
        "finetune_comment": finetune_comment,
        "trigger_word": trigger_word,
        "file_data": encoded_zip,
        "iterations": iterations,
        "mode": mode,
        "learning_rate": learning_rate,
        "captioning": captioning,
        "priority": priority,
        "lora_rank": lora_rank,
        "finetune_type": finetune_type,
    }

    response = requests.post(url, headers=headers, json=payload)
    try:
        response.raise_for_status()
        return response.json()
    except requests.exceptions.RequestException as e:
        raise requests.exceptions.RequestException(
            f"Finetune request failed:\n{str(e)}\n{response.content.decode()}"
        )


def finetune_progress(
    finetune_id,
    api_key=None,
):
    if api_key is None:
        if "BFL_API_KEY" not in os.environ:
            raise ValueError(
                "Provide your API key via --api_key or an environment variable BFL_API_KEY"
            )
        api_key = os.environ["BFL_API_KEY"]
    url = "https://api.us1.bfl.ai/v1/get_result"
    headers = {
        "Content-Type": "application/json",
        "X-Key": api_key,
    }
    payload = {
        "id": finetune_id,
    }

    response = requests.get(url, headers=headers, params=payload)
    try:
        response.raise_for_status()
        return response.json()
    except requests.exceptions.RequestException as e:
        raise requests.exceptions.RequestException(
            f"Finetune progress failed:\n{str(e)}\n{response.content.decode()}"
        )


def finetune_list(
    api_key=None,
):
    if api_key is None:
        if "BFL_API_KEY" not in os.environ:
            raise ValueError(
                "Provide your API key via --api_key or an environment variable BFL_API_KEY"
            )
        api_key = os.environ["BFL_API_KEY"]
    url = "https://api.us1.bfl.ai/v1/my_finetunes"
    headers = {
        "Content-Type": "application/json",
        "X-Key": api_key,
    }

    response = requests.get(url, headers=headers)
    try:
        response.raise_for_status()
        return response.json()
    except requests.exceptions.RequestException as e:
        raise requests.exceptions.RequestException(
            f"Finetune listing failed:\n{str(e)}\n{response.content.decode()}"
        )


def finetune_details(
    finetune_id,
    api_key=None,
):
    if api_key is None:
        if "BFL_API_KEY" not in os.environ:
            raise ValueError(
                "Provide your API key via --api_key or an environment variable BFL_API_KEY"
            )
        api_key = os.environ["BFL_API_KEY"]
    url = "https://api.us1.bfl.ai/v1/finetune_details"
    headers = {
        "Content-Type": "application/json",
        "X-Key": api_key,
    }
    payload = {
        "finetune_id": finetune_id,
    }

    response = requests.get(url, headers=headers, params=payload)
    try:
        response.raise_for_status()
        return response.json()
    except requests.exceptions.RequestException as e:
        raise requests.exceptions.RequestException(
            f"Finetune details failed:\n{str(e)}\n{response.content.decode()}"
        )


def finetune_delete(
    finetune_id,
    api_key=None,
):
    if api_key is None:
        if "BFL_API_KEY" not in os.environ:
            raise ValueError(
                "Provide your API key via --api_key or an environment variable BFL_API_KEY"
            )
        api_key = os.environ["BFL_API_KEY"]

    url = "https://api.us1.bfl.ai/v1/delete_finetune"
    headers = {
        "Content-Type": "application/json",
        "X-Key": api_key,
    }
    payload = {
        "finetune_id": finetune_id,
    }

    response = requests.post(url, headers=headers, json=payload)
    try:
        response.raise_for_status()
        return response.json()
    except requests.exceptions.RequestException as e:
        raise requests.exceptions.RequestException(
            f"Finetune deletion failed:\n{str(e)}\n{response.content.decode()}"
        )


def finetune_inference(
    finetune_id,
    finetune_strength=1.2,
    endpoint="flux-pro-1.1-ultra-finetuned",
    api_key=None,
    **kwargs,
):
    if api_key is None:
        if "BFL_API_KEY" not in os.environ:
            raise ValueError(
                "Provide your API key via --api_key or an environment variable BFL_API_KEY"
            )
        api_key = os.environ["BFL_API_KEY"]

    url = f"https://api.us1.bfl.ai/v1/{endpoint}"
    headers = {
        "Content-Type": "application/json",
        "X-Key": api_key,
    }
    payload = {
        "finetune_id": finetune_id,
        "finetune_strength": finetune_strength,
        **kwargs,
    }

    response = requests.post(url, headers=headers, json=payload)
    try:
        response.raise_for_status()
        return response.json()
    except requests.exceptions.RequestException as e:
        raise requests.exceptions.RequestException(
            f"Finetune inference failed:\n{str(e)}\n{response.content.decode()}"
        )


def get_inference(
    id,
    api_key=None,
):
    if api_key is None:
        if "BFL_API_KEY" not in os.environ:
            raise ValueError(
                "Provide your API key via --api_key or an environment variable BFL_API_KEY"
            )
        api_key = os.environ["BFL_API_KEY"]
    url = "https://api.us1.bfl.ai/v1/get_result"
    headers = {
        "Content-Type": "application/json",
        "X-Key": api_key,
    }
    payload = {
        "id": id,
    }

    response = requests.get(url, headers=headers, params=payload)
    try:
        response.raise_for_status()
        return response.json()
    except requests.exceptions.RequestException as e:
        raise requests.exceptions.RequestException(
            f"Inference retrieval failed:\n{str(e)}\n{response.content.decode()}"
        )


if __name__ == "__main__":
    import fire

    fire.Fire()

​
Best Practices and Tips

​
1. Enhancing Concept Representation

  • Try finetune_strength values >1 if the concept is not present
  • Increase the finetune_strength if the concept is not present or the identity preservation is not strong enough
  • Lower the finetune_strength if you are unable to generalize the concept into new settings or if the image has visible artifacts

​
2. Character Training

  • Avoid multiple characters in single images
  • Use manual captions when multiple characters are unavoidable
  • Consider disabling auto-captioning in complex scenes or for complex concepts

​
3. Quality Considerations

  • Use high-quality training images
  • Adjust learning rate based on training stability
  • Monitor training progress and adjust parameters as needed

​
4. Prompting

Change the trigger word to something more contextual than “TOK”. For example, “mycstm sunglasses” for a product finetune on sunglasses.
Key Strategies:
  • Prepend the trigger word to your prompt:
    "mycstm sunglasses, a photo of mycstm sunglasses laying on grass"
  • For character and product consistency, briefly describe the person/product:
    "mycstm man, a photograph of mycstm man, sitting on a park bench and smiling into the camera. mycstm man is a middle aged man with brown curly hair and a beard. He is wearing a cowboy hat."
  • For styles, append “in the style of [trigger-word]“:
    "a kangaroo wearing ski goggles holding up a drink, in the style of mycstm watercolourstyle"

​
Polling URLs for Finetuned Inference

When running inference on finetuned models, the same polling URL principles apply: 
import requests
import os
import time

# Submit finetuned inference request
response = requests.post(
    'https://api.us1.bfl.ai/v1/flux-pro-1.1-ultra-finetuned',  # Use appropriate region
    headers={
        'accept': 'application/json',
        'x-key': os.environ.get("BFL_API_KEY"),
        'Content-Type': 'application/json',
    },
    json={
        'prompt': 'A photo of TOK in a serene landscape',
        'finetune_id': 'your-finetune-id-here',
        'finetune_strength': 1.0,
        'aspect_ratio': '16:9'
    }
)

data = response.json()
request_id = data['id']
# For legacy endpoints, construct polling URL manually
polling_url = f"https://api.us1.bfl.ai/v1/get_result"

# Poll for results
while True:
    time.sleep(0.5)
    result = requests.get(
        polling_url,
        headers={
            'accept': 'application/json',
            'x-key': os.environ.get("BFL_API_KEY"),
        },
        params={'id': request_id}
    ).json()
    
    if result['status'] == 'Ready':
        print(f"Finetuned image ready: {result['result']['sample']}")
        break
    elif result['status'] in ['Error', 'Failed']:
        print(f"Generation failed: {result}")
        break