The development team of the image generation AI “FLUX” has released “FLUX Portrait Trainer”, a new trainer specialized in generating high-quality portraits. Features include finer details, eye highlights, and better expression of various facial sizes.
Train Flux LoRAs For Portraits | Training | AI Playground | fal.ai
https://fal.ai/models/fal-ai/flux-lora-portrait-trainer?ref=blog.fal.ai
Introducing the FLUX Portrait Trainer
https://blog.fal.ai/introducing-the-flux-portrait-trainer/
Below is an image generated using the same model in FLUX Portrait Trainer that was released as a sample. Fine details such as each hair on the body, the highlights reflected in the eyes, and the texture of the skin are expressed with high quality, and the development team said, “The picture looks like it’s jumping out of the screen.” I am.
The image below is an image generated with the prompt “kptz woman in a pitch black room. I single shaft of volumetric light slants across her right eye.” based on the dataset provided by CheeseAI. Although the instructions include “pitch black room,” the dataset only contained photos taken during the day. The image on the left is generated by a model trained with FLUX Portrait Trainer and is able to output the desired image, but the image generated by the model trained with the general-purpose trainer on the right cannot be made completely dark.
Similarly, below is an image generated with the same seed and same parameters with the prompt “kptz old woman”. The image generated by the model trained with the FLUX Portrait Trainer on the left is exactly as requested by the prompt, while the image generated by the model trained with the general-purpose trainer on the right is an image that matches the data used for training. and the prompt is ignored.
The development team explains that FLUX Portrait Trainer is a complement to a general-purpose trainer, and that depending on the application, a different trainer may work better.
Copy the title and URL of this article