This model is an actual fine-tune of Flux dev (No merged LoRA), tailored towards generating content with a focus on science fiction, and a cinematic feel. I have tried improving ??????? and I wouldn't say I've succeeded but I think there are improvements over the base model.

Unlike the base Flux model, which I feel tends to have a green tint, this Brainflux version leans more towards a blue hue. It also tends to produce much darker results in shadowed or low-light scenes, while brighter scenes remain unaffected.

Brainflux is the result of multiple failed attempts at fine-tuning the model and I have spent too much time and resources to not share the results with the community...so here it is. I have learned a lot along the way and I'm hopeful whenever time allows for a V2, there'll be significant improvements over V1.

THIS IS JUST THE UNET, PUT THE .SAFETENSORS in your ComfyUI/models/unet directory and use it like the other UNETs.

This model isn't necessarily an improvement over the original Flux; rather, it's an alternative that offers different outputs. It is far from perfect, and users should consider it a starting point—something to experiment with and build upon. Use this model at your own discretion, knowing it might not fit every project or preference, but it could be interesting for those exploring darker, more sci-fi-oriented content.

Training Details

{
  ...,
  "discrete_flow_shift": 3,
  "double_blocks_to_swap": 6,
  "flux_fused_backward_pass": false,
  "fp8_base": true,
  "fused_backward_pass": false,
  "fused_optimizer_groups": 0,
  "guidance_scale": 1,
  "learning_rate": 1e-05,
  "learning_rate_te": 0,
  "model_prediction_type": "raw",
  "optimizer": "Adafactor",
  "optimizer_args": "relative_step=False scale_parameter=False warmup_init=False",
  "timestep_sampling": "sigmoid",
  "train_text_encoder": false,
  ...
}