This LoRA is an early proof of concept for a Full Finetune that will be done in the future for FLUX.

*Need 5,000 more buzz to release the full fine tuning + Final Lora Models!

Thank you to those who have already donated buzz towards the project!

*Any help is appreciated. As you know, these models take alot of time, resources and energy. I want to release the best possible model & full fine tuned model as soon as buzz is raised.

COMPARISIONS BELOW (WITHOUT LORA VS WITH LORA)

(very early version test model. The official version will be trained on 3x the dataset / Steps)

• Fully Captioned Dataset (High Quality .PNG Images)

• Trained for 9,060 Steps

• Took around 2 days to fully train

• Trained on Lewds / surreal-like, realistic NSFW images.

• Resolution: 1024x1024

• Lora Weight: 1.0 (Suggested)

• Final Goal : Full Finetune on thousands of images

Where the model excels :

• Creating Lewds / Art (Similar to DALL-E / Cici Style)

• Pair with other LoRAs

• "Flash Photograph" Prompts

• Follows prompts well

• Multiple Aspect Ratios (Square, Landscape, Portrait)

• Gets rid off "Flux Face" when generating females

• "Facials", Body Shots, etc.

• Issues:

  • More images of genitalia can be added to the dataset

  • No images of Men in dataset

  • Hands are ok.

  • Larger dataset to give even more flexibility

  • Far away faces need some retuning

What Are The Next Steps?:

• Remove some watermarks out of the dataset, Release the Full LoRA Version (Trained with Positions / Acts + Everything)

• Release the Anime / Cartoon / Animation Version

• Release the Full Finetuned Version (Everything combined into one model)

Any buzz will go toward the Full LoRA / Full Finetune Training!

If you like the lora and can donate Buzz, its much appreciated!

Leave any feedback to help improve the full model.

Hide