Flux .1 dev - v.2This is a retrained version of Yvonne Craig's Batgirl, built on Flux 1.0 dev. I've completely remastered the model from scratch, significantly improving the resemblance while enhancing flexibility and accuracy. This version should offer a better balance between likeness, quality, and versatility compared to previous iterations.For best results, use the default Flux workflow with either euler or deis samplers, set between 20 and 50 steps. Keep the LoRa weight at 1 for optimal outputs.This model is part of my ongoing research and learning process, so I may release an improved version in the future. Feel free to experiment and please share your creations!Flux .1 dev - v.1This is my first LoRA model trained for Flux, using the same dataset I previously used for the SDXL version. I'm amazed at how well Flux generalizes and captures every detail. As this is just my initial experiment, I plan to iterate and upload a much-improved version soon.In the meantime, feel free to try it out and share your feedback if you'd like. You might notice that the facial resemblance is a bit off in this version. This is because Flux is highly sensitive and captures small details very well, which unfortunately includes some issues present in my original dataset. I wasn't able to fix the facial details in the source material back then, and this carried over to the new model.Don't worry, though! I now have a different approach in mind and will be addressing this issue soon. Have fun experimenting with the model!<hr />SDXLI wanted to train a high-quality model, but it was a difficult project because the source images were blurry. I had to restore the faces of the dataset one by one with low denoising inpainting trying not to lose the main features of Yvonne Craig. Some of her facial resemblance was lost in the restoration process, but the costume came out amazing. I think this is the best I can do given the quality of the source material. Even though her face is not 100% accurate because of this, I’m releasing it because I think it can produce some beautiful images. I hope you enjoy it!Context:Yvonne Craig was an actress who played Barbara Gordon, the daughter of Commissioner Gordon, in the 1966 Batman series. She also became Batgirl, a crime-fighting partner of Batman and Robin, in the third and final season of the show. She was known for her purple and yellow costume, her motorcycle, and her martial arts skills. She was one of the first female superheroes on television and a pioneer for the genre.Usage:Activation Words: yvonne craig womanRecommeded settings:LoRa Weight: 0.7. Please use this weight if you want to ensure te best likeness and output.Clip Skip: 1Sampler: All the DPM++ SDE work great. I personally prefer DPM++ 3M SDE KarrasResolution: at least 1024 x 1024 pxUse ADetailer if you want but try to not go over 0.4 denoising strenght.If your hardware allows it, give it a 1.5 hires fix with a denoising of 0.2 - 0.3 and you will get much more details. The model has a good amount of details on the costume.Upscaler: 4xNMKD-Superscale-SP_178000_G or 4xNMKD-Superscale-_150000_GI made the sample images using Automatic 1111 because for some reason CivitAI doesn’t embed the metadata of my ComfyUI images, but I already made a few generations there and it works the same. In fact, I use ComfyUI more than Automaitc 1111 these days.

Yvonne Craig Batgirl

This is my first LoRA model trained for Flux, using the same dataset I previously used for the SDXL version. I'm amazed at how well Flux generalizes and captures every detail. As this is just my initial experiment, I plan to iterate and upload a much-improved version soon.In the meantime, feel free to try it out and share your feedback if you'd like. You might notice that the facial resemblance is a bit off in this version. This is because Flux is highly sensitive and captures small details very well, which unfortunately includes some issues present in my original dataset. I wasn't able to fix the facial details in the source material back then, and this carried over to the new model.Don't worry, though! I now have a different approach in mind and will be addressing this issue soon. Have fun experimenting with the model!

Files with SHA256: 0b32e0ff45bccb61ad156689b10586c8e194ec7d6d8e2a5c09eea9500fa7230f