This article talks about tools like Astria.AI, and OpenArt.AI, which can be used to train AI models with faces that it doesn’t know. Then you can generate images with text prompts or use preset prompts from samples. I already wrote a blog post explaining how to use several Stable Diffusion based AI tools to create images using text prompts - https://www.bulbapp.io/p/375ad7b7-533c-427c-9874-3cf98782da51/ai-tools-for-text-to-image-generation.
We also discuss a couple of AI tools used for image-to-text generation. And give an example of how to use negative prompts.
Lastly, we introduce the use of ChatGPT for generating and manipulating text, as well as Lex, a word processor with AI built in.
To follow up on the text to image tools that I talked about, I thought I would mention 2 more tools that do image to text prompt generation (the reverse process). It works from any kind of digital image, or even a photograph. Here are 2 sites which allow you to do this:
clip_interrogator.ipynb - Colaboratory
Both of them do a pretty good job of analyzing the image and coming up with a good prompt. Of course, if you use the resulting prompt you won’t get the exact same image (which is to be expected). I find them useful to get ideas for new images since I don’t usually save the prompts, and seed numbers for most of my images.
The first one is pretty fast and easy. The second one takes a little longer, but is more accurate. It uses Google collab, which you might be familiar with.
I also wanted to talk about how to use “Dreambooth” tools below for training your own AI models and generating images from this process. You could use photos of yourself, someone you know personally, or a pet for example. You can also use different kinds of objects that the general models don’t know about, such as a new product that you are developing, for example.
This is a very useful tool which allows you to upload about 10-20 pictures of yourself or someone else, or even your dog. Then it will generate a bunch of sample images using its own pre-defined prompts, which can be modified, or you can use your own prompts. It is quite easy to use, the instructions are pretty good, and the results are very good if you follow the instructions closely. But you will have to use another tool (they recommend lexica) to generate more prompts. There is a small charge to get started, but I found it to be very worthwhile.
This tool has a similar process, is quite inexpensive and has everything you need on the same site. It will generate images based on “preset templates” which you can chose. You can also use the same site to find sample images and prompts to use on your pictures to generate even more. There is a quota of 400 images for each model that you train, which is more than sufficient. I tested it with photos of me, my girlfriend, and my grand daughter so far. Maybe I will use our dog for the next model. Its a lot of fun, and you can find a lot of uses for the pictures online as avatars, or PFP’s or whatever.
Here is another helpful hint for generating images from prompts using SD 2.1. It is a good idea to use negative prompts. Try it and see how the results will improve. For example:
"disfigured, kitsch, ugly, oversaturated, grain, low-res, Deformed, blurry, bad anatomy, disfigured, poorly drawn face, mutation, mutated, extra limb, ugly, poorly drawn hands, missing limb, blurry, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, ugly, disgusting, poorly drawn, childish, mutilated, , mangled, old, surreal"
(These are all things that you don't want to see in your images)
ChatGPT
Similar to generating and manipulating images, there are lots of AI tools for generating and manipulating text for things like blog posts or web site content. I have just started experimenting with ChatGPT, and I find it useful mostly for getting new ideas, or augmenting my current content rather than producing copy per se. You can experiment with the tool here: https://chat.openai.com/chat
Lex - A word processor with AI built in
This text was written with the help of Lex. Here is a video which explains how it works - https://www.youtube.com/watch?v=7Cao0oy1CBg
“Right now Lex uses GPT-3, so it can do pretty much anything GPT-3 can do which is quite a lot! For example:
Come up with what you should say next
Generate counter-arguments
Generate outlines for essays
Summarize text
Adopt a specific style (only works for well-known writers)”
It doesn’t seem to be be live yet, so let me know if you would like an invite.
I used Lex to generate the title, and also this conclusion.
“The tools mentioned above are just a few of the ones out there. There are many more and no doubt more are being created all the time. These tools open up new possibilities and make it easier to create content, whether it is images or text. They can also be used to improve existing content and create new ideas. If you have any questions or would like to know more, please feel free to contact me.”
I plan to continue writing blog posts about AI tools that I have been evaluating, and also some of the “next generation” social media platforms based on Web3 concepts.
Fred Blauer
