Tabarak Paracha
This video is part two of a trilogy discussing ‘questions about generative AI tools’ featuring Ioana Hreninciuc, Co-founder of PicFinder. PicFinder is a unique web-based AI software for text-to-image generation. The first part of this trilogy explored the question “How can AI simplify tasks for mobile developers?” and was released last week. You can watch it here.
In this video, we asked Ioana to answer the following questions:
How can we write a good prompt using AI?
Ioana, you recently co-founded an AI-tool called PicFinder. Could you tell us more about it?
“With PicFinder, we can make image generation as fast as image search. It looks very much like a search engine.
(The image above shows a screenshot of PicFinder’s main page)
We use stable diffusion models so (to write prompts most effectively) you’d want to follow a structure like the subject, the background or the context that your character is placed in, then you’d want to mention the style of the image that you’re trying to achieve, then you’d want to mention the lighting the character is placed in.
The order should be the following: the most important thing always goes first, and then in decreasing importance, you add the other elements.”
Could you show us how to write prompts in PicFinder with examples?
Here is an example of a prompt for creating a background:
“Empty living room, minimal, 3D render, illustration, soft lighting”
(The image above illustrates the prompt results generated from the prompt “Empty living room, minimal, 3D render, illustration, soft lighting.” The image aspect ratio chosen here is square)
The idea is to keep playing with the prompts by adding to the prompt until you have an image that you can use.
As an example, here’s what happens when I add “minimal” and “unfurnished” to the prompt:
(The image above illustrates the prompt results generated from the prompt “Empty living room, minimal, 3D render, illustration, soft lighting, minimal, unfurnished.” The image aspect ratio chosen here is square)
You can also use PicFinder to create characters. And you can also play around with different formats (portrait vs landscape etc).
Here is an example of how you would write a prompt for creating characters in PicFinder:
(Prompt used by Ioana): “Young woman with house in background, smiling, vivacious, minimal cartoon, 2d illustration, flat, animated”
(The image above illustrates the prompt results generated from the prompt “Young woman with house in background, smiling, vivacious, minimal cartoon, 2d illustration, flat, animated.” The image aspect ratio chosen here is square)
You can also test out the different models with PicFinder to see what results you get and you can get various generations depending on how specific you make your prompt.
(The image above illustrates the different styles and orientations available in PicFinder as of August 2023. They can be accessed under “Settings” on the top right of your screen)
(The image above illustrates the prompt results generated from the prompt “Young woman with house in background, smiling, vivacious, minimal cartoon, 2d illustration, flat, animated.” The image aspect ratio chosen here is portrait)
The prompts are fairly easy, the key is to start experimenting. You can change the prompts and figure out what you want by playing around with it. It is also possible to click an image and get more images in that style.
You can also upload images and get similar style pictures.”
Stay tuned for more on generative AI tools
This isn’t all we asked Ioana to share her expert opinion on. Stay tuned to learn about what Ioana thinks are the most common pitfalls for developers that use AI in part three of this video trilogy.
