Just use his name, "Henry Cavill", as part of your prompt. I also experimented by adding a couple of characters he portrayed to the LoRA, try "Henry Cavill in a textured blue and red Superman costume from DC Comics" or "Superman suit from DC Comics" as well as "Henry Cavill as Geralt of Rivia from The Witcher". No guarantees though.

I prefer .8 strength but it's tough to test locally since I use the nf4 quantized version of Flux with Forge (I have a puny 8GB VRAM card, that's not changing anytime soon). If you have questions, let me know. Otherwise, help me learn.

Training Details:

• 72 images, 20 repeats, 10 epochs

• Natural language captioning beginning with "a photograph of Henry Cavill"

• 1024 resolution with aspect ratio bucketing with a batch size of 4

• 16 network dim, 16 alpha

• LR of 0.0005

• Constant with warmup

• The cost was 2k buzz and some change using CivitAI's LoRA trainer

I'm not certain warmup worked tbh So this may be constant scheduler. I also doubt these are the best settings and I'll tinker with it in the future.