If anyone is interested, here is my journey to V4.

First of all, FLUX and a 3070 are really painful, and I'm already looking forward to upgrading my Loras with a 5090. I upgraded the V3 image set using Image-to-Image and various detailers. (At this point, I really wish people had six fingers, but I digress).

After the training, the results were somewhat disappointing; sometimes the details were there, sometimes they weren’t. But I think I’ve learned a lot from the metadata of other very good Loras.

Then I played around a bit with ComfyUI and, by chance, resized the Lora with a workflow, which somehow had a strong impact. More details were now visible, so I thought, why not merge my anime Loras at lower strength to see if it brings more details? Well, the result was a bit of a lottery whether you get anime or not. After that, I simply merged these Lora with all my Flux Loras at different strengths and kept making small adjustments.

It was really fun creating the images and seeing multiple Loras come together in one picture. Well, it's not perfect yet, and I’ll keep improving it, but for now, I had enough. I wish you all a lot of fun!