Description of Basic Functions of YES Ai Bot
YES Ai Bot is a system that combines access to multiple neural networks, each performing different tasks. Specifically, Chat GPT can generate text content and answer questions, while Stable Diffusion has virtually unlimited capabilities for creating images and videos.
In the bot https://t.me/yes_ai_bot, you have access to the main menu where you can manage your account: deposit your balance, withdraw funds earned through the affiliate program (in the /balance section), renew your subscription and upgrade your tariff plan (in the /tariffs section), and get your referral link to invite new users (in the /ref section).
The bottom menu is the main tool for managing the settings of the neural networks.
The last button switches between the Stable Diffusion neural network (for image and video generation) and the Chat GPT neural network (for text content generation).
In Chat GPT mode, two toggles are available: Remember Context and Increased Text Length.
Remember Context - This function allows the Chat GPT neural network to remember the dialogue history with you to consider the context. For example, if you ask the bot to "Write an article about raising children," it will send you a written article. Suppose you feel that the article miss some aspect. In that case, you can ask the bot to add information about teaching children to wash their hands before meals. The bot will understand that it needs to add to the article it previously sent to you.
If you disable Remember Context, the neural network will treat each of your requests as separate and unrelated to the previous conversation history. This mode of communication with the bot is suitable for getting answers to specific questions, such as "In which year was the city of Tokyo founded?"
Important! We recommend reading the article on how to write prompts (text prompts) correctly for the Chat GPT neural network. You can find the article at the following link: https://telegra.ph/Kak-pisat-prompty-dlya-Chat-GPT-06-18
This function allows you to partially bypass Chat GPT's limitations on the volume of generated text, which is useful for writing articles, long descriptions, etc. However, it is better to disable the increased text volume function when asking the bot to translate text from one language to another or when creating programs.
Stable diffusion settings are divided into two categories: SD settings and style settings.
The SD settings display the basic functions of the neural network: checkpoint selection, sampler selection, number of image generation steps, image resolution in pixels, CFG Scale (saturation), number of generations for each request, Denoising Strength (prompt influence), + GPT (SD and Chat GPT symbiosis), Negative Prompt.
Important! We recommend reading the article that explains how to write prompts (text queries) correctly for the Stable Diffusion neural network. Here is the link to the article: https://telegra.ph/Kak-pisat-prompty-dlya-Stable-Diffusion-06-18
In the style settings, you can choose one or several pre-prepared styles in which your images will be generated. This section also includes the LoRA gallery.
Now let's take a closer look at each option, starting with the basic SD settings.
The selected checkpoint is marked with a yellow marker - the neural network core is the most important part of the settings, as it is the core (checkpoint) that determines how the neural network will create images based on your text prompts. The checkpoint includes artificial intelligence knowledge about existing objects, how people, animals, objects, and everything else looks. It is important to note that the checkpoint determines the drawing style, in particular, the degree of realism, etc.
Reliberate (https://civitai.com/models/79754/reliberate) the checkpoint designed to create images with maximum realism. For example, if you are drawing people, the neural network will create a deep texture of the skin, fine wrinkles, and so on.
Deliberate (https://civitai.com/models/4823/deliberate) has slightly lower detail than Reliberate but still maintains a very high level of realism. This checkpoint is versatile and works well for both realistic textures and art.
RPG (https://civitai.com/models/8124/a-zovya-rpg-artist-tools) is a checkpoint that allows you to work with concept art, game characters, and more. This core is not intended for generating realistic images but rather for drawing "fantasy" objects.
DreamShaper (https://civitai.com/models/4384/dreamshaper) can create images in both realistic and art styles. The results are similar to Deliberate, but with a greater emphasis on drawing.
DarkSushi (https://civitai.com/models/24779) is an excellent checkpoint for drawing in an anime style.
Samplers are processors that determine how the neural network will interpret your text requests to generate images. Samplers have less impact on the results compared to checkpoints. Both the checkpoint and sampler must be selected simultaneously, as both parameters affect the generation results.
Euler a is a classic sampler designed for fast processing with a high level of detail.
DPM++ 2M Karras is a sampler that works well for creating art, anime, landscapes, and more.
DPM++ SDE Karras is a relatively "heavy" sampler that processes images slightly slower than the others but provides excellent detail work. It is suitable for increasing realism, detailing skin, and more. When using it, the probability of mutations (extra fingers, limbs, etc.) decreases.
UniPC (predictor sampler) works well with a large number of sampling steps, such as 35. This sampler is often used for creating art.
The number of generation steps determines the number of cycles the neural network will work on the generated images. The more steps, the more details the neural network will "draw." In most cases, 24 steps are sufficient, but if you decide to increase this value, it may result in excessive detailing.
Pixel resolution. Neural networks are not designed to create large images at the first step. Instead, they create different variations of images, and users select the best results to enlarge. This is why our bot creates images in relatively small resolutions.
You have access to three formats: square (640x640), portrait (640x840), and landscape (840x640). When using portrait or landscape orientation, mutations may occur (duplicate characters, limbs, heads, disproportionate growth, etc.). This is due to the nuances of artificial intelligence and is not a bug.
CFG Scale determines how strongly Stable Diffusion will be tied to the written text request. In other words, it is the strength of the prompt's influence on the generation.
Number of copies - a parameter that determines the number of images generated for each prompt (text description that you send to the bot).
Denoising Strength (DenS) - the strength of the prompt's influence when modifying images. This parameter is only used by the bot when increasing images and using the INPAINT function (drawing/correcting selected image elements). The higher the "DenS" value, the stronger the changes will be.
+ GPT - if this option is activated, text prompts sent to the bot in free form will be processed by the Chat GPT neural network. For example, "a girl sitting on a bench in the park in the rain holding a dog on a leash" is a free formulation, our bot will automatically detect this and give the task to Chat GPT to rephrase the request so that it can be correctly processed by the Stable Diffusion neural network. For reference, Stable Diffusion accepts requests in a certain format, i.e. you need to list the necessary objects separated by commas, indicate weights, and use the correct syntax, for example, "(girl sitting) on a bench, holding a dog on a leash, rain". The topic of prompting (creating correct requests) will be covered in a separate article.
If we summarize, the + GPT function allows combining the capabilities of the Stable Diffusion and Chat GPT neural networks to simplify the process of writing prompts. This is particularly useful for beginner users.
Negative Prompt (NP) - this option is designed to prevent the neural network from generating undesirable content. The default settings already include the NP prompt, which reduces the likelihood of mutations of fingers, hands, limbs, etc. However, you can modify or add your own prompts. For example, if you don't want the generated images to include the color pink, simply add the text "blue color, blue" to the NP prompt, separated by commas. Using the same principle, you can restrict the generation of extra elements, body parts, decorations, etc.
Now let's discuss the "Style Settings" option in the bottom menu.
We have already added a lot of popular styles that you can activate with a simple button press. Each style is a carefully crafted text prompt that is automatically added to your textual descriptions of generated images.
Styles greatly simplify the process of generating content. All you need to do is select the desired combinations of styles and choose them in the correct order. For example, if you want to draw a girl in an anime style but with a high key, you can press the "Anime" style button and then the "Light" button. That's it! You just need to send the bot a request describing how this girl should look.
The last button to discuss is the "LoRA Gallery," but this topic is quite extensive and deserves a more detailed explanation. To learn how to use the LoRA Gallery in the YESAiBot, please read here: https://telegra.ph/Kak-ispolzovat-galereyu-stilej-LoRA-06-18
You can find the full list of articles about the @yes_ai_bot bot at this link: https://telegra.ph/Menyu-statej-YES-Ai-BOT-06-18
For any questions, please contact our Telegram community at: https://t.me/yes_ai_chat