SPECIAL FEATURE The case for creative AI (Part 2)
Earlier this month Christian Darkin shared his research and insight into the rise of creative AI. Now, he's back with Part 2 to show us how we can embrace the new technology and use it in our work.
AI and how to use it
You can produce an artwork or poem by prompting an AI with a couple of words - just as you can create an abstract artwork by pouring some paint on a canvas. But we’re artists. We’re not interested in doing that. We have brave, amazing new ideas and we want tools that help us realise them. What we’re interested in is how we can use AI to make what we already do better, more interesting, and different. This requires precision, patience, skill and imagination, but it offers the possibility to create art which has been impossible up to now.
If you’re an artist
Illustration in children’s books is vibrant and creative and wonderful. However, we as artists are massively limited by budget and time. Many (not all) picture book images are watercolors. Artists have done wonderful work in this medium, but this choice is driven by the fact that watercolours are quick, and the materials are cheap.
Imagine children’s books illustrated in detailed oil paintings. Or imagine using the kind of art-photography only previously available in fashion magazines. How about collage where every piece of material or texture was generated individually in a style to suit the book?
All these are things that have never been possible before, simply because publishers don’t have the budgets. Well they’re possible now.
Maybe you’re trying out ideas for a picture-book character. Why not get an AI to produce 1000 variations on your original concept sketches to feed your imagination? Some will be dull and boring, some will be wild and strange, but somewhere, you’re sure to see some facial features, clothing, or a hair style - or just an expression that gives you an idea from which to build.
Looking to paint a fairy castle? The AI has seen all the castles in the world, why not get it to generate a few, adding in some prompts about atmosphere and style, and use them just as you’d use reference images you’d find on google.
Pick your tools
There are a lot of AI image generators around. Three are particularly important:
1) Dall-E: The first useful one to come out. It’s very easy to use, and pretty good. It’s expensive though, at around 10p per image (you’re likely to have to create dozens of images for each you use). There isn’t a lot of customisation available.
2) Midjourney: A bit trickier to use and a lot cheaper, Midjourney can create some wonderful images, but it does have its own style. You have to use it in Discord - which means everything you do is visible to everyone who wants to see it.
3) Stable diffusion: The choice for serious artists. It’s completely free and completely open source. You can download it to your own PC (if you have a good graphics card). It’s techy and hard to work with, but people all around the world are constantly creating add-ons for it which means it’s possible to have an amazing level of artistic control. You can also train it yourself on your own images, so you can create consistent characters, and allow it to understand your own visual style.
If you’re a writer
ChatGPT probably isn’t up to creating compelling prose right now, not without substantial line-by-line editing. However, when plotting, and world-building, it can be a fantastic tool.
Let’s say you’re writing a scene in which your protagonist is in a physics lesson. You’ve probably got great ideas about your main character, their backstory and their motivations. What about the kid sitting next to them in class? How thoroughly have you developed that character? Why not describe the scene to ChatGPT and get it to give you thirty descriptions and backstories for children in the same class? You don’t have to use any of them, but if anything sounds interesting, you can give that character back to the AI, add a few details and ask for a complete breakdown of how they might behave in any situation. I think of this as imaginative research - you’re creating a world, adding the details that are important to you, and then using AI to help you fill out the background with colourful detail.
You probably know a bit more about your main character, but you could probably know more. Why not tell an AI what you know about their past, their speech, their personality, and then ask the AI to respond as though it is your character. You can then have a chat with them. You can ask them questions, and the AI will try to answer. Why not put them in a situation - say the first day at school? Ask them how they’re feeling, get them to describe a specific event in detail. Again, you can use the ideas that come up as a jumping off point for your own imagination, or you can ignore them.
Stuck on a descriptive paragraph? You probably already go to google to search for a photo of a person, or a place to help you come up with compelling descriptions. Why not take it a step further and get Stable Diffusion to create a unique image of (say) a spooky forest, or a spaceship toilet, and use that to spark your descriptive imagination?
Plotting not your strong point? Why not give ChatGPT your synopsis, and get the AI to critique it? Get it to locate your inciting incident. Get it to break the story down into 25 suggested chapters. Get it to suggest subplots, or find the “save the cat” plot beats.
Ready to submit your work, but dreading writing your one-page story outline? ChatGPT is brilliant at summarising and writing emails. Give it whatever you have - a chapter breakdown, a rough set of notes, a few important paragraphs picked from the manuscript, and get the AI to rewrite it into a sparkling one page summary. It’ll still need some editing work, but at least you’ll have the basics.
Some technical advice to get you started:
- Learn prompting skills. To get an AI to create something, you have to ask it. The exact words you use have a massive effect on the outcome. Experiment to find what works for you, and keep checking online to see what prompts other people are using and what results they are getting. Prompt Writing is likely to become a skill as important as grammar or sketching in the future.
- ChatGPT is accessible and fun to play with. It’s great for getting to know the technology, but it’s very careful about giving misleading answers. This is not great for fiction. Try using GPT3 instead - it’s a lot more willing to be imaginative.
- Ask for a lot of suggestions - whether you’re creating pictures or text, you’ll reject a lot of what the AI produces, so go for quantity. Say, “Give me 10 ways a ghost might communicate with the living” , “List 20 possible characters who might live in a hippy commune in 1968”, “Give me 10 emotional responses to the death of a pet.”
- Turn notes into sentences. Give the AI a few notes, and it can flesh them out into a sentence. This is good for expanding and clarifying your ideas.
- Best practice is changing all the time. Two months ago, it was difficult to reproduce consistent characters in AI art. Now it's possible. Last month, getting specific poses was a challenge. Now it’s a lot more possible. Find AI artists on YouTube and follow the developments. In this fast moving field.
- Tell the AI it’s good. This may sound strange, but the prompt, “The following is the plot of an award winning classic children’s story written by a master storyteller” will get you better results than “Write a children’s story”. Likewise, adding “Beautiful, imaginative professional illustration” to a prompt for a picture will tell the AI you don’t just want clipart.
What’s next?
So what’s next for creative AI? I start every day now by checking what’s possible today that wasn’t possible yesterday. When new ideas and techniques hit, they appear on YouTube first. I check the following people to see what’s new:
https://www.youtube.com/@MattVidPro
https://www.youtube.com/@DavidShapiroAutomator
https://www.youtube.com/@TwoMinutePapers
https://www.youtube.com/@sebastiankamph
https://www.youtube.com/@DrAlanDThompson
Last year was the year of image and language generation. 2023 is likely to see this technology become more controllable for artists. We’ll have more and better ways to edit our AI work, and finetune it. Writing AI will allow more context (i.e. we’ll be able to ask it to consider more detailed background - an entire draft chapter of our book, for example - when responding). We’re likely to see AI being able to search the internet too. Search will become a lot more chatty.
By the end of the year, you’ll be able to type in text to generate a video. Eventually, it’ll be possible to create the film of your book using only the manuscript and your own creative directions.
It’s certain that there will be an explosion of poorly created AI books on Amazon, and that agents will be inundated with low quality work. We as artists are used to this kind of competition. It’s also certain there will be an explosion in very high quality work that uses the new technology in innovative and creative ways.
Readers, and publishers will have to find ways to become more discerning. Probably this too will involve AI - In the future we may well be getting an AI to “pre-read” books before we order them to let us know if we’re going to like them - or at least whether they reach a basic level of literacy.
In the meantime, go out and start experimenting. The first Carnegie prizewinning book co-written with artificial intelligence is yet to be written. I'd argue that this is primarily because nobody has figured out how to prompt AI to help write it. In the next year or so, somebody may, and it could be you.
*Header: Christian's test image using AI. All images: courtesy of Christian Darkin
*
Christian Darkin is an author, illustrator and animator who has been following the rise in AI since writing about it two decades ago for The Times and The Guardian. In a two part series, he presents a (possibly controversial) positive view of the present and future of this technology.
*
Stephanie Cotela is the Network News & Events Editor at Words & Pictures magazine.
I can't agree with Christian about AI. I've had a little play on Chat GPT and it's fun to see what it can (and can't) do. But then you remember that it's theft, especially the way AI manipulates images. And that's without the evidence from the likes of Bradford Literature Festival who have already cut out the middle man (i.e. us). They didn't employ an illustrator to produce promotional material using AI to make it spectacular and exciting. They just went straight to the AI and got it for free. Not a great look for a lit fest. We need to take a long hard look at the consequences. For many SCBWI members that means they'll be out of a job. To think anything else is naive in my opinion.
ReplyDeletedf
ReplyDeleteAnother tip is ensuring the content is not difficult to peruse yet in addition as connecting as could be expected, as this implies that imminent clients are bound to understand it and divide between their companions.
ReplyDeleteChatGPT Prompts