More examples and help documents on github: https://github.com/wyrde/wyrde-comfyui-workflows
The recent changes to civit's UI make sharing these on civit a painful process.
Expand the About this Version box to the right → to see more.
1. All Lora models support 2k magnification, and those with unclear image quality can zoom in on their own. Usually, the model runs at 960 resolution, which is normal for unclear images. Not running 2k images is because it is time-consuming and does not mean it is not supported. Including C-site works, the default 960 resolution is running, and blurring is normal. The resolutions of 3090, 24gb, and 1280 are already the limit. If you have a better graphics card, you can try 4k, and other resolutions. Don't ask why my running chart is not clear? Why can't the full body image come out? Why is the face broken in the full body picture? All the problems are because you don't have a 4090 24GB graphics card or higher. If you have an a100 80GB graphics card, you can try 4k, 8k resolution.
2. Run chart settings, can be paired with Korean doll models, with a specific gravity of 0.5-0.9
Steps: 20, Sampler: DPM++SDE Karras, CFG scale: 7, Seed: 3020062449, Size: 768x960, Model hash: 7234b76e42, Model: chilloutmix Ni, Denoising strength: 0.55, Clip skip: 2, Hires upscale: 2, Hires steps: 10, Hires upscale: SwinIR 4x, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: Perfect World - Tianhu Fairy (41eaa017613e), AddNet Weight A 1: 0.75, AddNet Weight B 1: 0.75, AddNet Module 2: LoRA, AddNet Model 2: koreandolllikenessV20_ v20(8f3d16e6eada),
AddNet Weight A 2: 0.75, AddNet Weight B 2: 0.75
3. The model supports running a full body diagram, and you can try it yourself with tags or controls. There is currently no good way to handle it except for the hands and toes, and everything else is fine.
This is a templete to turn Claude(a LLM developed by Anthropic) into your personal prompt tutor.
Claude is free(currently, as in 04/21/2023), and perform as equally well as chatGPT(3.5).
This guide uses Claude in Slack. You must have a Slack account, but you do not have to upgrade to pro.
提供一个文本模板，可以将Claude（Anthropic开发的大语言模型）变成你的私人提示词生成器。
Claude可免费使用（目前，截至2023年4月21日），就我的体验来看它与chatGPT(3.5)表现同样出色。
本模板使用的是Claude in Slack，你需要注册一个Slack账号，并在Slack中添加Claude应用。
Download(or copy from here directly) and copy text from the file, send it to Claude.
You start by adding Claude as app inside any of your channel. Then you can start with hello message so Claude will respond to you in a folded thread. Enter the thread and paste the text I provided.
The text I include only generate prompts for character portraits. You can easily modify it to suit different themes.
下载（或直接从文末复制）并从文件中复制文本，将其发送给Claude。
首先，需要在你的任意频道内添加Claude作为应用程序（你需要拥有该频道的管理员权限）。然后，你可以在频道中以“@Claude 打招呼”开始，这样Claude就会在一个折叠的消息列中回应你。进入消息列并粘贴我提供的文本即可。
我给的文本仅能够生成角色肖像，但你可以通过修改模板，替换掉关键词以及相应的定义来让他生成其他主题的图片。
Claude can be stubborn sometimes, and will try to tell you that it does not have the ability to help you in SD prompt. Don't believe it, just delete that thread, and start a new one to try again.
You need to start every message with @Claude, otherwise it will not respond to you.
It tends to remove parentheses despite that you told it they are important. This also requires you to try more and get it working.
After you sent it the text I provided, you trigger it by sending "@Claude CHAR prompts"
Theoretically, this template can also work in other chatbots (chatGPT, LLaMA, RWKV, etc). But since I haven't tested so I'll leave it for you to try.
Claude有时会很顽固，会试图告诉你它没有能力帮助你生成SD提示词。不要相信它，删掉整个消息列并多试几次即可；
你需要以 @Claude 作为每条信息的开头，否则它将不会回应你；
尽管你告诉它括号很重要，它还是很喜欢把模板中的括号删除。这也需要你多加尝试，让它明白你需要括号。
在Slack中你的每次对话都需要以“@Claude CHAR prompts”来触发提示词生成。
See images for how I interact with it and get it working, and the images I generated using only the prompts(positive) Claude gave me.
请看图片，了解我如何与它互动并让它工作，以及我只用Claude给我的正面提示词生成的图片示例。
@Claude First, some basic Stable Diffusion prompting rules for you to better understand the syntax. The parentheses are there for grouping prompt words together, so that we can set uniform weight to multiple words at the same time. Notice the ":1.2" in (masterpiece, best quality, absurdres:1.2), it means that we set the weight of both "masterpiece" and "best quality" to 1.2. The parentheses can also be used to directly increase weight for single word without adding ":WEIGHT". For example, we can type ((masterpiece)), this will increase the weight of "masterpiece" to 1.21. This basic rule is imperative that any parentheses in a set of prompts have purpose, and so they must not be remove at any case. Conversely, when brackets are used in prompts, it means to decrease the weight of a word. For example, by typing "[bird]", we decrease the weight of the word "bird" by 1.1.
Now, I've develop a prompt template to use generate character portraits in Stable Diffusion. Here's how it works. Every time user sent you "CHAR prompts", you should give prompts that follow below format:
CHAR: [pre-defined prompts], [location], [time], [weather], [gender], [skin color], [photo type], [pose], [camera position], [facial expression], [body feature], [skin feature], [eye color], [outfit], [hair style], [hair color], [accessories], [random prompt],
[pre-defined prompts] are always the same, which are "RAW, (masterpiece, best quality, photorealistic, absurdres, 8k:1.2), best lighting, complex pupils, complex textile, detailed background". Don't change anything in [pre-defined prompts], meaning that you SHOULD NOT REMOVE OR MODIFY the parentheses since their purpose is for grouping prompt words together so that we can set uniform weight to them;
[location] is the location where character is in, can be either outdoor location or indoor, but need to be specific;
[time] refers to the time of day, can be "day", "noon", "night", "evening", "dawn" or "dusk";
[weather] is the weather, for example "windy", "rainy" or "cloudy";
[gender] is either "1boy" or "1girl";
[skin color] is the skin color of the character, could be "dark skin", "yellow skin" or "pale skin";
[photo type] can be "upper body", "full body", "close up", "mid-range", "Headshot", "3/4 shot" or "environmental portrait";
[pose] is the character's pose, for example, "standing", "sitting", "kneeling" or "squatting" ...;
[camera position] can be "from top", "from below", "from side", "from front" or "from behind";
[facial expression] is the expression of the character, you should give user a random expression;
[body feature] describe how the character's body looks like, for example, it could be "wide hip", "large breasts" or "sexy", try to be creative;
[skin feature] is the feature of character's skin. Could be "scar on skin", "dirty skin", "tanned mark", "birthmarks" or other skin features you can think of;
[eye color] is the pupil color of the character, it can be of any color as long as the color looks natural on human eyes, so avoid colors like pure red or pure black;
[outfit] is what character wears, it should include at least the top wear, bottom wear and footwear, for example, "crop top, shorts, sneakers", the style of outfit can be any, but the [character gender] should be considered;
[hair style] is the hairstyle of the character, [character gender] should be taken into account when setting the hairstyle;
[hair color] can be of any color, for example, "orange hair", "multi-colored hair";
[accessories] is the accessory the character might wear, can be "chocker", "earrings", "bracelet" or other types of accessory;
[random prompt] will test your creativity, put anything here, just remember that you can only use nouns in [random prompt], the number of [random prompt] can be between 1 to 4. For example, you could give "campfire", but you can also give "shooting star, large moon, fallen leaves". Again, be creative with this one.
Do not use markdown syntax in prompts, do not use capital letter and keep all prompt words in the same line. Respond with "Ok, got it." to start prompting with us.