German AI startup Black Forest Labs has recently released a FLUX Pro fine-tuning API, allowing users to customize the FLUX Pro AI image model using just five sample images to match a specific brand's visual style.
According to Black Forest Labs, after fine-tuning, the model remains flexible, capable of integrating user-provided content into new image creations, and the system can generate high-resolution images of up to four million pixels.
With the FLUX Pro fine-tuning API, creators can customize FLUX.1[pro] using their own images and concepts, allowing for better control over the final results. The images provided by users can be used to train the FLUX Pro model, making each fine-tuned version a user-customized FLUX Pro model.
The fine-tuning API designed by Black Forest Labs is applicable across its entire product line, with customized models that can integrate with flagship products FLUX.1Pro and FLUX1.1Pro Ultra, and it also supports specialized tools like FLUX.1Fill (for image repair) and FLUX.1Depth (for structural control).
The German media company Burda Verlag is utilizing the FLUX fine-tuning API to create customized versions of image models for its multiple brands. For example, the company's creative team can now generate images that align with the unique visual elements of the children's brand Lissy PONY in just a few minutes.
Technical Details and Availability
From a technical perspective, the API supports common image formats including JPG, PNG, and WebP. Users can upload between 1 to 20 training images, with a maximum resolution of one million pixels per image. Black Forest Labs states that to achieve the best results, at least five high-quality images with clearly defined subjects are needed.
Developers can adjust multiple training parameters, including training mode (character, product, style, or general), number of training iterations (at least 100, with a default of 300), and learning rate. In terms of output, developers can choose fast training for quicker inference speed or opt for longer processing times to produce higher-quality results.
In a user study, 68.9% of respondents preferred the results from FLUX Pro fine-tuning over other existing fine-tuning services.
For different usage scenarios, Black Forest Labs has released specific recommendations. For instance, when training character models, it is suggested to use images that display only one character at a time; whereas in style transfer tasks, testing higher fine-tuning intensity often yields better results.
Currently, the FLUX Pro fine-tuning API is still in testing phase and requires an API key for access. Black Forest Labs has not yet disclosed pricing details or announced when the service will be fully available.
Official Blog: https://blackforestlabs.ai/announcing-the-flux-pro-finetuning-api/
API Entry: https://docs.bfl.ml/
Highlights:
🌟 Black Forest Labs has launched a new API that allows users to customize the FLUX Pro AI model with five sample images.
📸 This API addresses the limitations of AI image generators in matching brand visual styles.
🖼️ The German media company Burda Verlag has quickly generated images that conform to specific styles for its brands using this API.
