Detail Tweaker XL_chauhanWhat it isDetail Tweaker XL_chauhan is a slider LoRA. It does not change the subject. The strength number is the control: positive weight adds texture, negative weight flattens it. There is notrigger word.Same prompt, weight only. Left to right: 0, -1.0, 0.6, 1.0, 1.5.InstallComfyUI: put the .safetensors file in ComfyUI/models/loras, refresh, and add a Load LoRA node between the checkpoint and the sampler. Set strength_model. Forge orAutomatic1111: put it in models/Lora and write in the prompt.UseGenerate once with the LoRA off and keep that seed. Rerun the same seed at -1, 0.6, and 1. Do not add a trigger. Start at 0.6. Use it on the SDXL-family checkpoint it was trained on.On another family it will only add noise. Output size is not locked to 512.Weight Effect-2 to -1 Flatter, smoother, simpler texture-0.5 Slightly softer than the base0 Base model. LoRA has no effect0.4 to 0.8 Useful detail: pores, weave, hair strands1.0 Strong detail, still mostly safe1.5 and above Crunchy. Lower the weight or use an earlier fileIf the face changes at 0.6, the weight is too high for that save. Drop toward 0.4, or use an earlier training step.

Detail Tweaker XL_chauhan

Detail Tweaker XL_chauhanWhat it isDetail Tweaker XL_chauhan is a slider LoRA. It does not change the subject. The strength number is the control: positive weight adds texture, negative weight flattens it. There is notrigger word.Same prompt, weight only. Left to right: 0, -1.0, 0.6, 1.0, 1.5.InstallComfyUI: put the .safetensors file in ComfyUI/models/loras, refresh, and add a Load LoRA node between the checkpoint and the sampler. Set strength_model. Forge orAutomatic1111: put it in models/Lora and write in the prompt.UseGenerate once with the LoRA off and keep that seed. Rerun the same seed at -1, 0.6, and 1. Do not add a trigger. Start at 0.6. Use it on the SDXL-family checkpoint it was trained on.On another family it will only add noise. Output size is not locked to 512.Weight Effect-2 to -1 Flatter, smoother, simpler texture-0.5 Slightly softer than the base0 Base model. LoRA has no effect0.4 to 0.8 Useful detail: pores, weave, hair strands1.0 Strong detail, still mostly safe1.5 and above Crunchy. Lower the weight or use an earlier fileIf the face changes at 0.6, the weight is too high for that save. Drop toward 0.4, or use an earlier training step.

Files with SHA256: 0ec2ec31bfdf5cf06afa82d7df41449ce728ca8718d276638f0e58f3b54a8d0e