stable diffusion webui之lora调用

1.触发词+底模+lora效果最好(分数不一定要取到1,0.8也行);

2.引用时一定要使用,例如

stable diffusion webui之lora调用_第1张图片

"prompt": "(masterpiece:1.3), (best quality:1.3), (no human), artwork of 3D geometry, still life, clean, Pastel color, Comfortable natural light, minimalist style, modernist style, 3d rendering, 8K,",

"negative_prompt": "False , unreal , drawing , lines , low quality , lowresolution , blurry , unclear,nsfw"

stable diffusion webui之lora调用_第2张图片

lora在webui中调用链路: 

modules/processing->process_images->

res = process_images_inner(p)->

711
p.parse_extra_network_prompts()->
- extra_networks->parse_prompts()->parse_prompt()

713
with devices.autocast():
    extra_networks.activate(p,p.extra_network_data)
- extensions-builtin->Lora->extra_networks_lora->ExtraNetworkLora.activate->
- lora.load_loras(name,multipliers)->
- lora.py->load_loras()->
-- lora_on_disk = loras_on_disks()->
-- lora = load_lora(name,lora_on_disk)->

你可能感兴趣的:(多模态和生成模型实践,stable,diffusion)