Luis Royo Flux

New version for Flux !

The example images were made with fluxDevQ5KM (I don't have a big PC), euler simple, 28 steps, guidance 3.5, lora force 1.0

It is better to use something like "Painting of ..."

Feel free to like if you like ;)

Training settings, if it can help fine-tuners:

{
  "steps": 2516,
  "lora_rank": 16,
  "optimizer": "adamw8bit",
  "batch_size": 1,
  "resolution": "512,768,1024",
  "autocaption": false,
  "trigger_word": "TOK",
  "learning_rate": 0.00015,
  "wandb_project": "flux_train_replicate",
  "wandb_save_interval": 100,
  "caption_dropout_rate": 0.05,
  "cache_latents_to_disk": false,
  "wandb_sample_interval": 100
}

For image capture I used "llava-13b" with the question "Give me a detailed description of this image?", then deleted what did not explicitly describe the image.