Trained with 29 photos (8 face, 10 half body, 11 full body) - 10 epochs. Base Realistic Vision V6.0 B1.
**Trigger words: redwitch, mikaela reid dbd
Working config: 512x768, Steps: 6, CFG: 1.5, Sampler: DPM++ SDE Karras, Checkpoint: RealisticVisionV60B1_v51HyperVAE.safetensors, Weight: <lora:redwitch:0.7>
💡 My personal recommendation:
I trained and tested this LoRA with Realistic Vision V6.0 B1 HyperVAE, and that's where I like it the most.
That's where the skin, freckles and curly hair look the most natural, more like the real Mikaela.
If you try it with normal Realistic Vision (not HyperVAE) it also works, just at 25 Steps, CFG 7 it looks a little more polished, more like a studio photo. Personally I prefer the 6 steps version because it looks more natural and softer.
If you use it at 25 steps I recommend lowering the weight a little to 0.6 so it doesn't look too perfect: <lora:redwitch:0.6>