Stable Diffusion 换装图

Stable Diffusion换装秀、AI画中国洛神图 

让人物在保持现有姿势不变的情况下对小姐姐的图片进行换装。 

选取了《洛神赋》里面的洛神, 让Stable Diffusion重绘中国传世名画。

AI换装秀 

1. 准备换装图片: 首先,先准备一段 tag (prompt), 我们用它生成一个AI小姐姐的图片。当然,你也可以自己准备一张现实中拍的图片。

2. 局部重绘操作:点击右下角的图生图按钮,进入图生图操作界面,然后选择局部重绘。 

     

3. 设置蒙板 : 对小姐姐的衣服进行换装,需要更换的只是衣着,其他地方包括姿势要保持不变,所以我们用蒙板将要换装的部分全部涂黑,画的时候仔细一点,尽量不要多画。

为了换装的边界过滤更顺滑,可以将下方的蒙版模糊值拉大一点,设为10就差不多了。

因为要让衣服生产随机变化,所以我们将随机种子调为1,其它参数保持不变。

注意:图片尺寸设为和原始图片尺寸一致的512*768。

4. 启用ControlNet 

  • 点击下方的ControlNet,点击启用按钮。
  • 预处理器和模型均选择OpenPose。
  • 注意,图片尺寸要保持一致,设为512*768。

5. 更换衣服关键词: 在关键词中将原有的衣服关键词换掉,我这里写了一个白色T恤,你也可以将白色T恤的关键词加权重,比如我要将它增加1.3倍的权得,可以这样写:(White T-shirt:1.3) 

一切设置完毕,直接点击生成即可。生成的图片并不全是白色T恤,不过没关系,而且其实AI已经帮我们实现了我们想要的换装操作 。

知识点 

如果想更换不同风格的衣服,可以在提示词中加入不同风格的衣服,中间用“|”分隔,然后用中手中号将它们括起来,代表这些衣服将随机出现。 

文章来源“青橙ing” 2023-04-03

重绘《洛神赋》里面的洛神 

采用了保存得最好的北京故宫博物院甲本《洛神赋图》作为原画。为了排除干扰,对原画进行了处理,换了一个干净一点的底纹,去掉了树枝,只保留人物。为了检验AI能否智能化去噪点,没有对人物面部和身上的杂点进行处理。 

启用了控制网络中的Canny插件,以保持出图动作的一致性。 

咒语倒是很简单:best quality,1girl,looking down, beautiful face,tangstyle,longuette,standing on the lake,holding a fan in hand,straps on clothes,whirl about in the wind,coldly elegant,, < lora:koreanDollLikeness_v10:0.4>, 

尝试了几次,最好的结果是这样: 

于是更换模型,给予Stable Diffusion更多的自主权,然后AI就给惊喜了: 

一番尝试下来,发现看来AI不能很好的处理原画上的杂点,导致生成的画上出现了许多点状线条。 

文章来源“一笑聊知” 2023-04-09

山水风景 

pic 1

模型和咒语:

prompt: masterpiece, best quality, high quality, extremely detailed CG unity 8k wallpaper, scenery, outdoors, sky, cloud, day, no humans, mountain, landscape, water, tree, blue sky, waterfall, cliff, nature, lake, river, cloudy sky,award winning photography, Bokeh, Depth of Field, HDR, bloom, Chromatic Aberration ,Photorealistic,extremely detailed, trending on artstation, trending on CGsociety, Intricate, High Detail, dramatic, 

Negative prompt: worst quality, low quality, medium quality, deleted, lowres, comic, (watermarks)

Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 1760969915, Size: 512x1024, Model hash: fedb2f1fbe, Model: 3Guofeng3_v32Light, Clip skip: 2, ENSD: 31337 

模特 

model

Checkpoint 大模型:SHMILY油画风_v2.1

Lora模型:自然景观_V0.2,场景画_v1,水彩插画_v1.0

外挂VAE: vae-ft-mse-840000-ema-pruned

prompt: On the intersecting streets of light and shadow, the demeanor of the female sports model is like a moving painting, captivating. She stands tall and elegant in the center, like a bamboo that has weathered storms yet remains indomitable.

The gleam in her eyes conveys an ineffable passion for sports, a yearning for self-challenge. When she immerses herself in movement, those eyes seem to ignite, shining with resolute light, guiding her body through precise and powerful motions one after another.

A slight upward turn of her lips forms a calm smile, acknowledging her own efforts and filled with boundless hopes for the future. This smile, like the warm sunshine of spring, is gentle yet powerful, compelling people to approach, feeling the serenity and confidence from within.

In the demeanor of the female sports model, we see her love and dedication to sports, and we sense her inner strength and tenderness. Every subtle expression seems to tell a story of courage, persistence, and dreams, moving people deeply.

Negative prompt: The image is fuzzy, not high-definition, with impurities and spots, and overexposed.

Steps: 26, Sampler:DPM++ 3M SDE Karras, Size:1024X1086

真人模型 

点击《Stable Diffusion 人脸特美的人像摄影大模型》文章 (作者:跟着安琪学AI)