Stable Diffusion - ReV Animated v1.2.2 的 2.5D 模型与提示词

欢迎关注我的CSDN：https://spike.blog.csdn.net/
本文地址：https://spike.blog.csdn.net/article/details/131655644

ReV Animated v1.2.2 模型能够创建类似 2.5D 的图像生成。该模型是 checkpoint 合并，是其他模型的产物，以创建源自原始模型的产品。

VAE：

Orangemix.vae.pt (推荐)
kl-f8-anime2.ckpt
Blessed2.vae.pt

Prompting:

Order matters - words near the front of your prompt are weighted more heavily than the things in the back of your prompt.
Prompt order - content type > description > style > composition (内容类型>描述>风格>构图)
This model likes: ((best quality)), ((masterpiece)), (detailed) in beginning of prompt if you want anime-2.5D type
This model does great on PORTRAITS

Negative Prompt Embeddings:

EasyNegative
Deep Negative
bad_prompt_version2
bad-artist
bad-artist-anime
bad-quality
Make use of weights in negative prompts (i.e (worst quality, low quality:1.4))

启动 SD 服务：

conda deactivate
source venv/bin/activate
# python launch.py --port 9301 --xformers
CUDA_VISIBLE_DEVICES="1" nohup python -u launch.py --port 9301 --xformers --no-half-vae --theme dark > nohup.sd.out &

模型运行

测试模型：https://civitai.com/models/7371/rev-animated

提示词 outdoor 提示会导致人物的拿包站立姿态，如果不拿包，建议去除，直接使用 street 。
负面提示词 pointy ears,cat ears,(horns:1.2)，去除人物头上的尖角。

提示词：

(full body:1.2),1girl,cyberpubk girl,a rebellious girl,a girl with a tattoo,
beautiful eyes,charming smile,earrings,makeup,long hair,standing postures,
colored pigment,colored paint,colorful,
street,alone,
official art,unity 8k wallpaper,ultra detailed,beautiful and aesthetic,beautiful,masterpiece,best quality,
<lora:ClothingAdjuster2:-0.2>,<lora:Cute_girl_mix4:0.3>,<lora:3DMM_V12:0.45>,<lora:tutu_cyberpunk_v5:0.25>,
Negative prompt: (logo:1.2),(watermark:1.2),username,patreon username,patreon logo,pointy ears,extra fingers,deformed hands,(polydactyl:1.5),(worst quality, low quality:1.4),(monochrome),zombie,badhandv4,bad_prompt_version2-neg,EasyNegative,
Steps: 20, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Seed: 1916158388, Face restoration: CodeFormer, Size: 512x768, Model hash: 4199bcdd14, Model: RevAnimated_v122, Denoising strength: 0.4, Hires upscale: 2, Hires steps: 30, Hires upscaler: 4x-UltraSharp, Lora hashes: "ClothingAdjuster2: f038e3a5b67b, Cute_girl_mix4: 4768d15b1b67, 3DMM_V12: 7d5c705f0ff8, tutu_cyberpunk_v5: fe183c895d19", Version: v1.4.0

图像参考：

去除尖尖的耳朵的效果，提示词：

official art,unity 8k wallpaper,ultra detailed,beautiful and aesthetic,beautiful,masterpiece,best quality,
(full body:1.2),1girl,cyberpubk girl,a rebellious girl,a girl with a tattoo,
beautiful eyes,charming smile,earrings,makeup,long hair,standing postures,
colored pigment,colored paint,colorful,
street,alone,
<lora:ClothingAdjuster2:-0.2>,<lora:Cute_girl_mix4:0.3>,<lora:3DMM_V12:0.45>,<lora:tutu_cyberpunk_v5:0.25>,
Negative prompt: EasyNegative,badhandv4,bad_prompt_version2-neg,(logo:1.2),(watermark:1.2),username,patreon username,patreon logo,pointy ears,cat ears,(horns:1.2),,extra fingers,deformed hands,(polydactyl:1.5),(worst quality, low quality:1.4),(monochrome),zombie,
Steps: 20, Sampler: DPM++ 2M SDE Karras, CFG scale: 7, Seed: 1916158388, Face restoration: CodeFormer, Size: 512x768, Model hash: 4199bcdd14, Model: RevAnimated_v122, Denoising strength: 0.4, Hires upscale: 2, Hires steps: 30, Hires upscaler: 4x-UltraSharp, Lora hashes: "ClothingAdjuster2: f038e3a5b67b, Cute_girl_mix4: 4768d15b1b67, 3DMM_V12: 7d5c705f0ff8, tutu_cyberpunk_v5: fe183c895d19", Version: v1.4.0

图像：

LoRA & Embeddings

位置：

Lora: stable-diffusion-webui/models/Lora
Embeddings: stable-diffusion-webui/embeddings/

LoRA: ClothingAdjuster2 衣物增减

参考：Clothing +/- Adjuster 衣物增/减 LoRA，推荐：<lora:ClothingAdjuster2:-0.2>，负值是增加衣物。

介绍：这是一个调节所绘制对象衣物多少的功能性LoRA。通过将LoRA权重从-1.0调节至1.0，可以实现绘制对象衣物的逐步减少。本模型在生成竖图时效果最好。

LoRA: Cute_girl_mix4 少女脸

参考：Cute_girl_mix4，推荐：<lora:cuteGirlMix4_v10:(recommend0.4-0.7 here)>

介绍：work with Chilloutmix, can generate natural, cute, girls.

LoRA: 3DMM_V12 3D渲染脸型

参考：3D rendering style，配合模型 ReV Animated，推荐：<lora:3DMM_V12:0.45>

建议提示词：

1girl,Look straight ahead, long hair, Straight hair, a dark room with lights, close-up shot, Street Background, masterpiece, best quality,3dmm style,3d,realistic, high-quality, 3D realistic works, ultra-high details, 4k, real skin materials,simple background,Negative prompt: tattooing,lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry,braid hair

范例使用 Additional Networks 插件来调用 LoRA。

LoRA: tutu’s cyberpunk 机械朋克

参考：tutu’s cyberpunk，推荐：<lora:tutu_cyberpunk:0.25>

例如，无论你使用什么底片，如果你将（full body）的权重提高到例如1.7以上，那么很有可能会显示出全身图像。使用这个版本时，重量可以相对稍大一些。0.65是一个临界点。不同的权重有不同的效果。

Embeddings: bad_prompt_version2-neg 负面提示词

参考：bad_prompt Negative Embedding，参考：bad_prompt_version2-neg

这种嵌入背后的想法是以某种方式将否定提示训练为嵌入，从而将否定提示的基础统一为一个单词或嵌入。事实证明，嵌入对于手的生成非常有帮助！