Press "Enter" to skip to content

Posts published in “AI / ChatGPT”

(转载) Stable Diffusion绘画创意文字实操教程

AIGC技术不断更新迭代,国内出现了越来越多的新玩法,比如最近大家都在热议的AI绘画创意文字。

过去的一周,我把这些新玩法都研究了一遍,并总结了一套完整的制作流程。

主流的创意文字玩法主要分为四类:

  1. 光影文字
  2. 嵌入文字
  3. 隐藏文字
  4. 海报文字

每种类型的文字制作流程大致都是这四步

制作文字底图

选取大模型

写关键词(Prompt)

ControlNet参数设置

核心:文生图+文字底图(ControlNet)

本教程用到的所有资料,我都已经整理好放在文章末尾的网盘链接里,你不需要花费很多时间和精力去寻找了。

我在整理这些内容上花了很多时间,希望它们能对你有所帮助。

下面直奔主题:

通用部分实操

制作文字底图

美图秀秀,PS都可以,我这里使用的是不专业的美图秀秀

海报文字、嵌入文字用白底黑字

隐藏文字、光影文字用黑底白字

打开美图秀秀=》图片美化=》制作壁纸=》选择背景颜色=》下一步(进行文字编辑,不过多介绍)

选取大模型:

真人图片就用: majicmixRealistic,chilloutmix

其他用: revAnimated,realisticVision,deliberate

本文中所用到的所有模型资料,都会给大家打包整理好,放在文末的网盘链接里

有精力和时间的,大家可以多尝试,或许会抽到意想不到的效果。

1.光影文字实操:

1.1 制作底图(黑底白字):

这个图好像有点吓人

1.2 选取大模型:

选真实感大模型(majicRealistic或者chilloutmix)

1.3 写关键词(Prompt):

城市建筑通用词:

Best quality, ultra high definition, masterpiece, ultimate details, 8K,

Urban night view, buildings, cars, headlights, streets, roads, water reflection

最好的质量,超高清,杰作,极致细节,8K,

城市夜景,建筑,汽车,前灯,街道,道路,水反射

人物通用词:

The highest quality, masterpiece, ultimate detail,

1 girl, beautiful, full body, long skirt, standing with hands in pockets

最高的质量,杰作,极致细节,

1个女孩,漂亮,全身,长裙,站着,双手插在口袋里

StableDiffusion的关键词分类查询

①先写画面质量

②然后写小姐姐的外型描述

③最好加上一些“阳光”、“光和阴影”、“明暗对比”之类的词,并且加一点权重

sunlight, light and shadow, chiaroscuro

正面关键词:

((masterpiece)),(highres:1.3), (((best quality))), ((ultra-detailed)), ((illustration)),(photorealistic:1.3),(realistic:1.4),(Cinematic Lighting:1.3),

aynamic angle,in the winter,(snow white forest:1.1), hazy mist,,(on the hot springs:1.3) ,(many thin detailed snowflakes:1.3), (bloom), (shine), (twinkling stars:), (snowing), many white tree, detailed snowflakes,

cowboy shot,

(1 girl),(((solo))),running,(between legs:1.3),

(chinese girl:1.1),

hanfu,hanfugirl, beautiful detailed eyes,(black eyes:1.2),,(pale firered hair:1.3),light smile,

(big breasts:1.1),

,(beautiful detailed face:1.3),(cute), blushing,, light smile, makeup ,lipstick,detailed clothes,,(pale red transparent tight floral qipao dress:1.35),

(white fishnet thighhighs:1.25).

lora:add_detail:0.5,

sunlight, light and shadow, chiaroscuro

负面关键词:

(cleavage:1.4),(Multiple people),(((simple background))),monochrome ,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, lowres, bad anatomy, bad hands, text, error, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, ugly,pregnant,vore,duplicate,morbid,mut ilated,tran nsexual, hermaphrodite,long neck,mutated hands,poorly drawn hands,poorly drawn face,mutation,deformed,blurry,bad anatomy,bad proportions,malformed limbs,extra limbs,cloned face,disfigured,gross proportions, (((missing arms))),(((missing legs))), (((extra arms))),(((extra legs))),pubic hair, plump,bad legs,error,legs,username,blurry,bad feet ,(cleavage:1.4)

1.4 重要的ControlNet参数设置

我用的是tile模型

需要特别注意的就是控制权重(Control Weight)引导终止时机(Ending Control Step)

控制权重(Control Weight) :设置在0.35~0.55之间

引导终止时机(Ending Control Step) :0.7~1,越接近 1 字越清晰

没有固定的参数,不同的字需要分别调整,只能自己多试试

1.5 最终效果:

2.嵌入文字实操:

2.1 制作底图(白底黑字):

2.2 选取大模型

这里选择的是:deliberate,其他的revAnimated,realisticVision 也可以尝试

2.3 写关键词(Prompt)

梅花、红色、花瓣、冬天、雪白背景、古老树干、明亮蓝天、坚韧、生动、活力

Plum Blossom, Red, Petals, Winter, Snowy Backdrop, Ancient Trunk, Bright Blue Sky, Resilient, Vibrant, Dynamic

正面关键词:

Best quality, ultra high definition, masterpiece, ultimate details, 8K,

Plum Blossom, Red, Petals, Winter, Snowy Backdrop, Ancient Trunk, Bright Blue Sky, Resilient, Vibrant, Dynamic

负面关键词:

我这里找了一个关于景色的负面词:

(normal quality), (low quality), (worst quality), paintings, sketches,fog,signature,soft, blurry,drawing,sketch, poor quality, uply text,type, word, logo, pixelated, low resolution.,saturated,high contrast, oversharpened,dirt,

2.4 重要的ControlNet参数设置

接着就是controlnet的选择,这里就要选 canny,lineart,softedg,scribble 这些线稿的模型

在我实际测试下来,canny和scribble出来的效果都还可以,大家可以优先尝试这两个模型

降低权重,这样文字才能更好和场景融合到一起

可以把权重设置到 0.65~0.85 之间,想要文字更加突出权重就加大

2.5 最终效果

3.隐藏文字实操:

黑底白字

3.1 制作底图(黑底白字):

3.2 选取大模型

我这里选择的realisticVision,其他的也可以尝试:deliberate,revAnimated

3.3 写关键词(Prompt):

关键词: 英文:Oriental Pearl Tower, Dazzling Light, Night Sky, Spherical Structure, Skyline, Skyscrapers, Huangpu River, Bright Pearl, Shanghai.

中文:东方明珠、炫目光芒、夜空、球体结构、天际线、摩天大楼、黄浦江、璀璨明珠、上海。

正面关键词:

Best quality, ultra high definition, masterpiece, ultimate details, 8K,

Urban night view, buildings, cars, headlights, streets, roads, water reflection

Oriental Pearl Tower, Dazzling Light, Night Sky, Spherical Structure, Skyline, Skyscrapers, Huangpu River, Bright Pearl, Shanghai

负面关键词:

(normal quality), (low quality), (worst quality), paintings, sketches,fog,signature,soft, blurry,drawing,sketch, poor quality, uply text,type, word, logo, pixelated, low resolution.,saturated,high contrast, oversharpened,dirt,

3.4 重要的ControlNet参数设置

注意:controlnet的参数设置比较重要

这里我们需要下载一个新的controlnet模型

模型可以在文末的网盘下载

模型存放位置:sd目录 \ models \ ControlNet

不需要预处理器,模型就选我们新下载的

特别要注意的是控制权重,只要在0.45~0.65之间就可以了

控制权重可以多来回试试效果,需要字清晰一点权重就大一点

3.5 最终效果

效果不是太好,可以慢慢的调节参数。

4.海报文字实操:

4.1 制作底图(白底黑字):

4.2 选取大模型

我这里选择的是:revAnimated

4.3 写关键词(Prompt):

描述: 这幅画描绘了一池生机勃勃的荷花。在画面中心,一朵荷花在阳光下展开,粉红色的花瓣像丝绸般光滑,散发着淡淡的清香。绿色的荷叶如伞般展开,浮在宁静的池水上。几只蜻蜓在荷叶间穿梭,增添了一份夏日的热烈。远处,是微风吹动的芦苇和淡蓝色的天空,构成了完美的背景。

关键词: 英文:Lotus, Pink Petals, Sunshine, Green Leaves, Pond, Dragonflies, Reeds, Light Blue Sky, Serene, Summer.

中文:荷花、粉红色花瓣、阳光、绿色荷叶、池塘、蜻蜓、芦苇、淡蓝色天空、宁静、夏日。

正面关键词:

Best quality, ultra high definition, masterpiece, ultimate details, 8K,

Lotus, Pink Petals, Sunshine, Green Leaves, Pond, Dragonflies, Reeds, Light Blue Sky, Serene, Summer,

负面关键词:

我这里依旧使用上面的关键词:

(normal quality), (low quality), (worst quality), paintings, sketches,fog,signature,soft, blurry,drawing,sketch, poor quality, uply text,type, word, logo, pixelated, low resolution.,saturated,high contrast, oversharpened,dirt,

4.4 重要的ControlNet参数设置

预处理器选invert

模型选depth

知识补充

如果是生成液体材质的文字,比如牛奶,可以用canny,lineart,softedg,scribble这几个线稿

然后稍微降低一点controlnet的控制权重(0.6~0.8之间)

这样出来的文字就会有液体飞溅的感觉

4.5 最终效果


Methods for Creating Real AI Portrait from Stable Diffusion (text-to-image face swap)

AI portraits have gained popularity recently, especially with the emergence of a certain phenomenal camera. By simply uploading your photo, you can generate beautiful portrait pictures. This product has once again ignited interest in AI art. Today, I will share a method for creating real AI portraits using the Stable Diffusion WebUI, allowing for quick image generation without the need for training. Specifically, this method involves utilizing the face-swapping capability of Stable Diffusion’s Wensheng Image and the roop plugin.

Please note that this article only shares the technology behind creating AI portraits and does not encourage any malicious use. The author is not responsible for any face-swapping behavior conducted by readers after reading this article.

Installation of Plugins:

Two installation methods are introduced: one through the plugin extension and the other through manual uploading.

After the installation, you will see this option in the text-to-image or image-to-image tab, right below section of the Stable Diffusion WebUI

Method One:

Suitable for users with smooth access to various resources.

  1. Launch Stable Diffusion WebUI and go to “Extension Plugins” > “Install from URL.”
  2. Enter the following URL: github.com/s0md3v/sd-w…
  3. Click “Install.”
  4. Wait for the prompt “Use Installed tab to restart” to appear.
  5. Go to the “Installed” tab and click “Apply and restart the user interface.”


Note: if the webpage still fails to load after restarting for more than 5 minutes, check the console or terminal logs. It might be due to the slow download speed of the models required for the plugin. If the download is taking too long, you can interrupt the download and manually upload these files. For specific instructions on how to do this, refer to Method Two.

Method Two:

Suitable for users with restricted access to Github and HuggingFace.

You will need following files:

  1. Download the necessary files manually from Github and HuggingFace or use the provided compressed package:
  2. Deploy the files to their respective locations on the server/PC(self-host):

copy sd-webui-roop folder to stable-diffusion-webui/extensions/

copy models folder to stable diffusion root directory:

ifnude and insightface folder are a bit tricky; they need to be placed in the user’s root directory, and the folders should be renamed starting with an period.

Taking ifnude as an example,

For Windows: C:\Users{YourLoginUsername}.ifnude

For Linux: /root/.ifnude

For Linux systems, after uploading ifnude and insightface to the root directory, you can quickly rename them using the following command.

mv /root/ifnude /root/.ifnude 

mv /root/insightface /root/.insightface

At the end, restart Stable Diffusion WebUI

Face Swapping:

I will start by using ‘Text to Image’ to generate a photo of a 30-year-old Chinese man. The parameters below are just for your reference; as long as you can generate a portrait image, it’s sufficient. However, if you are new to SD (Stable Diffusion), feel free to follow the settings in the image entirely. Get the initial result and then customize it as needed.

The specific parameters for image generation are as follows:

Checkpoint: realisticVisionV20 (Download link available at the end of the article)

Prompt: ultra quality, masterpiece, best quality, 8k, UHD, a man, Chinese, standing in front of a desk, 30 years old, thin face, long hair, slight smile, (looking at viewer), upper body, solo, portrait photo, with a bookshelf behind, sense of technology, wide shot

Negative Prompt: EasyNegative, bad hands, bad fingers, glasses, worst quality, low quality, normal quality, skin spots, acne, age spots, ugly, deformed

expand roop plug in:

And then upload a new face image with following setting, you test using CodeFormer vs GFPGAN :

Finally click “Generate” button, after few minutes, you should see the result:

For image-to-image face swap:

In image-to-image tab, after upload the photo you want to be swap, then use followin setting:

in roop plugin, upload the new face:

If you only want to replace a specific area on the face, you can also use the inpaint tab, and use your mouse to paint those area. The operation method is shown in the following figure:

After add a mask to the desired facial area, keep other parameters unchanged, and click ‘Generate.’

Done!

Please note that this article only shares the technology behind creating AI portraits and does not encourage any malicious use. The author is not responsible for any face-swapping behavior conducted by readers after reading this article.

Prodia is a tool for generating AI photo stickers online for free, making it easy to create beautiful and personalized pictures

After sharing the teaching of Stable Diffusion to generate realistic human-like pictures before, if you want to simplify the steps, now there is a free Prodia platform, combined with a drawing function similar to Stable Diffusion, you can directly open the web page and generate it with one click, whether it is a mobile phone or a computer. Support, quickly generate personalized photo stickers online through Prodia.

Prodia online AI face painting generator

Prodia is a free AI drawing tool that integrates the Stable Diffusion 1.4 interface and 6 style models. It can also be used with Prompt commands to quickly generate various AI-style photos. It can be used without registration or installing any App program, which is convenient and can be downloaded directly picture.

Although the highest resolution of the free version of Prodia can only produce 512 X 512 pixel AI pictures, it is basically enough to be used as a profile picture or shared on social platforms, unless you want higher quality or excellent pictures, you should consider paying. It is enough to say that most of them are free!

Prodia teaching: Use AI drawing to quickly generate beautiful pictures with one click

After opening the Prodia platform through the browser (link at the end of the article), you will see the upper part similar to the Stable Diffusion setting interface. The first menu on the left can select the model, and the right can enter different commands and values, and finally press “Generate” directly. Immediately produce pictures. If you don’t know how to generate, the fastest way is to directly click the six different sample styles below, including hand-painted, realistic, realistic characters, CG animation style, and electronic drawing.

Prodia teaching: Use AI drawing to quickly generate beautiful pictures with one click

If you want to customize the module and parameter details, you can click the “Control Settings” icon to customize the Prompt, Negative prompt, Smapler, CFG Scale, Steps and Speed ​​values. Usually, you only need to set the module, Prompt, and Negative prompt Three types are fine, and the others do not need to be modified and adjusted. Below are some examples to show you how to use Prodia to generate beautiful AI images.

Prodia teaching: Use AI drawing to quickly generate beautiful pictures with one click 1

SD 1.4 Models: sketch, hand-painted, 3D style

SD 1.4 is more inclined to draw hand-painted and cartoon-type model styles. If you want to produce hand-painted sketch pictures, you can use this model to generate them. However, if you want it to draw a portrait of Luffy, some micro-faces or squinted eyes may appear in the final drawing.

Prodia teaching: Use AI drawing to quickly generate beautiful pictures with one click 2

Anything V3 model: CG style (Japanese)

The Anything V3 model is mainly biased towards Japanese CG character styles. You can try to use the Prompt command below to generate a CG killer-style female appearance.

lucy \(cyberpunk\), 1girl, arm up, asymmetrical hair, belt, bodysuit, covered mouth, covered navel, detached sleeves, grey eyes, hip vent, holding, holding weapon, looking at viewer, night, night sky, pouch, short hair, sky, solo, weapon, white hair, wire, short shorts, shorts, open jacket, <lora:lucy-000035:0.6>
Prodia teaching: Use AI drawing to quickly generate beautiful pictures with one click 3

Anything V4.5 model: CG style (European and American)

Compared with V3, Anything V4.5 also adopts European and American CG styles, but this group of models is more soft and less rigid. You can try to use the commands below to experience the style.

1girl, rogue,hood,(large breasts:1),skindentation, ,black cape , extremely detailed,harness,realistic(masterpiece, best quality:1.2) <lora:animeknight_v2:1>
lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, (worst quality, low quality, normal quality:1.2), jpeg artifacts, signature, watermark, username, blurry, EasyNegative,loli
Prodia teaching: Use AI drawing to quickly generate beautiful pictures with one click 4

Analog V1 Model: Real Life Looks

If you want to generate real-life AI photos, you can directly select the Analog V1 model to generate them. You must use the Prompt command to generate them. For the first time, you can try to use the examples of real-life girls in Examples to generate beautiful pictures of Asian girls. It is also possible. However, the probability of failure is slightly higher. This module is more biased towards Europe and the United States, and Asian female faces are prone to appear.

Prodia teaching: Use AI drawing to quickly generate beautiful pictures with one click 5

TheAlly’s Mix II Digital Portrait (Soft)

If you want to generate 3D electric-style character pictures, you can use TheAlly’s Mix II module, which will be more suitable. You can try it yourself by attaching the prompt command below.

(extremely detailed 8k ), (masterpiece), (best quality), (ultra-detailed), (best illustration), (best shadow), (photorealistic:1.4),(PureErosFace_V1:0.5),real human skin, lens flare, shade, backlighting, depth of field, natural lighting, hard focus, film grain, in front 1 girl, 21 yo, solo, skinny, (oversized clothes, t shirt:1.4), no panties, pale skin, soft lips, (light eyebrow:1.4), brown eyes, black very long straight hair, (small breasts), in classroom(crowd:1.2), aliga, a woman, , <lora:aliga_v1:0.8>
(worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), skin spots, acnes, skin blemishes, age spot, (outdoor:1.6), manboobs, backlight,(ugly:1.331), (duplicate:1.331), (morbid:1.21), (mutilated:1.21), (tranny:1.331), mutated hands, (poorly drawn hands:1.331), blurry, (bad anatomy:1.21), (bad proportions:1.331), extra limbs, (disfigured:1.331), (more than 2 nipples:1.331), (missing arms:1.331), (extra legs:1.331), (fused fingers:1.61051), (too many fingers:1.61051), (unclear eyes:1.331), bad hands, missing fingers, extra digit, (futa:1.1), bad body, NG_DeepNegative_V1_75T,pubic hair, glans
Prodia teaching: Use AI drawing to quickly generate beautiful pictures with one click 6

The bottom is an elf-style female style picture, and each produced work will present a very beautiful effect.

<(masterpiece, realistic:1.3), (extremely intricate:1.2)>, portrait of a girl, face, close up, pointy ears, dress, half-closed eyes, jewelry, sitting, strapless, strapless dress, breasts, watermark, bare shoulders, tiara, grey dress, cleavage, long hair, braid, grey hair, long eyelashes, solo, profile, solo, elf,
(worst quality, low quality:1.4), (depth of field, blurry:1.2), (greyscale, monochrome:1.1), 3D face, cropped, lowres, text, jpeg artifacts, signature, watermark, username, blurry, artist name, trademark, watermark, title, multiple view, Reference sheet, curvy, plump, fat, muscular female, strabismus,
prodia theallys mix ii a

Elldreth’s Vivid model: video game CG style portrait (shadow)

The last group of Elldreth’s Vivid models are more Korean video game CG style, with more obvious shadow effects, and the picture is closer to electronic drawing. As for the example, you can continue to use the above elf female style code.

Prodia Teaching: Use AI drawing to quickly generate beautiful pictures with one click 7

How to quickly apply Prodia commands?

After reading this Prodia tutorial, you will definitely want to produce other real-life photos with different styles. It is recommended to refer to the Civitai website. Most of the above pictures are produced using Stable Diffusion, and you can find what you want from there. Prodia photo style.

How to quickly apply Stable Diffusion works and image data?

After choosing the picture style you like, there will also be Prompt and Negative prompt descriptions inside, just paste these two descriptions on the Prodia platform.

stable diffusion google colab 19 1

function summary

If you want to quickly generate AI portrait photos through a computer or mobile phone, it doesn’t take too much time. It is highly recommended to use Prodia directly. The speed is relatively fast. If you want to produce more detailed and realistic portrait pictures, Prodia is not suitable. Mainly The reason is that the module is not fine enough, and there is no way to produce complex pictures of real people. It is easier to recommend Stable Diffusion. For detailed operation skills, please refer to “Free AI Real-life Drawing Teaching, Teaching You 4 Steps to Generate Realistic Beauty Photos “.