Getting AI to generate a picture you want takes practice" - Toky
Generative AI tools can create almost anything you can imagine. But how can you get what’s in your head onto a screen? Gareth Elis is joined by Toky and Ffion to find out.
GARETH: Our local festival is coming up and I'm running the merch stand. T-shirts, bucket hats, badges, posters… But I need ideas to make the store look cool.
Text-to-image generative AI tools can create almost anything you can imagine. They use large quantities of image data to generate new images from your prompts.
So, how can we use this for my merch stand and get what's in my head onto the screen? Toky and Ffion, I'm going to need your help.
TOKY: I'll start simple. I'm going to use a text prompt. 'Share inspiration images of a festival merch stand which will look cool and attract customers.'
FFION: And I'm going to use a visual image prompt. 'I took this photo last year at a festival and I think it would be a good starting point.'
GARETH: The text-only prompt means the AI uses natural language processing. That technology allows computers to generate human-like responses. For the image prompt, the AI uses computer vision, the technology that allows computers to see what's in our images and videos. So, let's see what we get.
TOKY: This looks weird. It looks really old-fashioned, and all the people look the same. And what's with the floating hat and T-shirt? To inspire us, I'm going to have to get more detailed.
FFION: This isn't like my picture at all. There's only one T-shirt, and it's all jewellery for sale. And it looks pretty empty.
GARETH: Okay. When AI receives a generic prompt, it defaults to the most statistically common patterns from its training data. Being specific helps you get more diverse and relevant outputs.
TOKY: I'm going to make my prompt more specific. We want to appeal to other teenagers, so I'll mention them. I'm also going to include bright colours and a neon sign.
FFION: I'll improve mine, too. The image can be a starting point, but I'll prompt the AI on what I want to be similar and what I want to be different.
TOKY: That's great. It's beginning to look more exciting.
FFION: Okay, that's looking good now. I'd buy something.
GARETH: Think before uploading personal photos. Always check the tool's privacy policy to see where your data goes and whether your images are stored or used to improve the generative AI. And remember, the generative AI tool isn't thinking creatively. It's executing your instructions. It doesn't have its own artistic intent. That part is unique to you.
FFION: Generative AI can be a bit predictable. It's up to us to create prompts that guide the AI to make something better.
TOKY: Getting AI to generate a picture you want takes practice. You have to try different prompt styles, follow up and change your instructions.
Text-to-image generative AI essentials
- You can use AI tools to generate images to help you be creative. Those images might inspire your own ideas or help you visualise what something could look like.
- If you use a text-only prompt, the AI uses natural language processing, technology that allows computers to understand what we say and generate a response.
- If you use a prompt with an image, the AI uses computer vision, technology that allows computers to 'see' and identify objects in images or videos.
- A generic prompt means a simple prompt that doesn’t give much detail or context. When an AI tool receive a generic prompt, it defaults to the most statistically common patterns from its training data.
- As AI tool training data is created by humans and reflects human biases, generic prompts often produce stereotypical results. If you input more detailed and specific prompts, you get more diverse and relevant outputs.
Tips for image generation with AI tools
| Do | Don't |
|---|---|
| Remember the generative AI tool isn’t thinking creatively, it’s executing your instructions. To get better results, try tweaking your prompts to guide it towards what you want to create. | Use vague or ambiguous prompts. And make sure you check the output - does it make visual sense? |
| Think before uploading personal photos or artwork. Check the tool's privacy policy to see where your data goes and whether your images are stored or used to train the generative AI tool. | Upload anything that you wouldn't share on social media. Never upload photos of anyone or their work without their permission. |
Quick guide: Writing text-to-image prompts
Think of what kind of image you want to generate and give clear and precise instructions.
| Subject: | Who or what is in the picture? | Person, animal, object |
| Location: | Where is it? | Indoors, outdoors, in the country, in the city, seaside, in the sky, underwater |
| Action: | What is happening in the picture? | Movement, still life |
| Form: | In what form? | Photo, painting, illustration, animation, sculpture |
| Colour: | What colours? | Colourful, monochrome, black and white, pastel shades, neon |
| Composition: | How is it framed? | Portrait, landscape, close-up, birds-eye view, fish-eye |
You can be even more precise by thinking about colours, medium and style.
| Medium: | If it’s a painting, what medium? | Oil painting, watercolour painting, pastel, pencil drawing, sketch |
| Style: | If it’s a painting, what style? | Abstract, figurative, photorealistic, impressionistic, expressionistic |
| Lighting: | If it’s a photo or painting, what lighting? | Daylight, sunny, overcast, stormy, backlit |
