ZedClampet
Community Contributor
Rewritten and expanded notes on local image generation:
Use Forge UI, which you can get from github or pinokio. This is easy to use and powerful. You can also try comfyui, obtained the same way, but it has a steeper learning curve. All of these things are free.
If you plan on using LoRAs (special training to make particular types of pictures) and you are using Flux, then you need the Dev version because no one makes LoRAs for the streamlined Schell version.
If you don't have at least an RTX 4080 GPU, use the "Diffusion in Low Bits" setting. Either of the "automatic" settings is fine, but the "bnb" settings were actually made by the developers of Forge, so they may provide a small speed boost.
The size of the picture you are asking for makes a huge difference in both speed and content. Make sure there is enough space for the AI to draw what you want. Also, you can have the AI make perfect upscales for you, so going too big could be a waste of time since upscaling is much faster than image generation.
These work off of natural language. Don't just throw out single words. It won't know what you want. Adding words like "detailed" or even something like "exciting" can help generate better pictures.
People generation
Many pictures will probably have people in them, but if you just want one person, you need to be exact in your description especially regarding age. If you want a woman mixing cookie dough, don't casually replace "woman" with "girl" because you will end up with a child. If you say "woman" but aren't more specific, you'll get a young woman in her mid 20s. If you want a little boy playing with a ball and you say "young boy", then you'll get anything from a 2 year-old to a 17-year old because all boys are young, so it just picks a random age. It can be difficult because the AI doesn't understand numbers, so saying "5-year-old" doesn't mean anything to it. The same problems apply to anyone over 30. Everyone over 30 is "old", so that doesn't help, and it can be very challenging to get an elderly person.
AI associates various things and acts accordingly whether you want it to or not. For instance, adding weight to someone generally makes them older, as does shortening a woman's hair. Hair length plays a huge role in what the AI thinks the face will look like, and a man with long hair is likely to look like a dashing pirate, while a young woman with short hair will often look like a punk drug user. And without great effort, everyone in the picture will be in the same mood. A little girl laughing as she holds a frog? Everyone else will be laughing too. There are just a lot of oddities you'll learn along the way when trying to get AI to make people.
I'm quitting now because I'm hungry, but there are many fascinating things about the images AI makes.
Use Forge UI, which you can get from github or pinokio. This is easy to use and powerful. You can also try comfyui, obtained the same way, but it has a steeper learning curve. All of these things are free.
If you plan on using LoRAs (special training to make particular types of pictures) and you are using Flux, then you need the Dev version because no one makes LoRAs for the streamlined Schell version.
If you don't have at least an RTX 4080 GPU, use the "Diffusion in Low Bits" setting. Either of the "automatic" settings is fine, but the "bnb" settings were actually made by the developers of Forge, so they may provide a small speed boost.
The size of the picture you are asking for makes a huge difference in both speed and content. Make sure there is enough space for the AI to draw what you want. Also, you can have the AI make perfect upscales for you, so going too big could be a waste of time since upscaling is much faster than image generation.
These work off of natural language. Don't just throw out single words. It won't know what you want. Adding words like "detailed" or even something like "exciting" can help generate better pictures.
People generation
Many pictures will probably have people in them, but if you just want one person, you need to be exact in your description especially regarding age. If you want a woman mixing cookie dough, don't casually replace "woman" with "girl" because you will end up with a child. If you say "woman" but aren't more specific, you'll get a young woman in her mid 20s. If you want a little boy playing with a ball and you say "young boy", then you'll get anything from a 2 year-old to a 17-year old because all boys are young, so it just picks a random age. It can be difficult because the AI doesn't understand numbers, so saying "5-year-old" doesn't mean anything to it. The same problems apply to anyone over 30. Everyone over 30 is "old", so that doesn't help, and it can be very challenging to get an elderly person.
AI associates various things and acts accordingly whether you want it to or not. For instance, adding weight to someone generally makes them older, as does shortening a woman's hair. Hair length plays a huge role in what the AI thinks the face will look like, and a man with long hair is likely to look like a dashing pirate, while a young woman with short hair will often look like a punk drug user. And without great effort, everyone in the picture will be in the same mood. A little girl laughing as she holds a frog? Everyone else will be laughing too. There are just a lot of oddities you'll learn along the way when trying to get AI to make people.
I'm quitting now because I'm hungry, but there are many fascinating things about the images AI makes.
Facebook
Twitter
Instagram