A picture is worth a thousand words. But with generative AI, you’ll need to start with words to get that image.
Generative artificial intelligence tech has exploded in the past few years, leading to a lot of creative services that can generate entire images and video clips from simple text descriptions called prompts. Programs like Canva make it easy to take advantage of the new AI tech for image creation and editing, no coding experience required. But there are a lot of AI tools and they aren’t all the same. I should know, since I’ve tested all of the major AI image and video generators available and then some. I’ve learned a lot about generative media while reviewing these AIs, and this guide is the culmination of every pro tip and experience-tested advice I have to give.
The TL;DR is that while these services are sometimes impressive and constantly improving, they’re not good enough to replace human creators. But if you want to experiment with what’s available, whip up a quick custom image or dive deeper into AI, this is everything you need to know about creating AI images.
Don’t miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred Google source.
Step 1: Pick the best AI service for your project
I’ve had hands-on experience with many AI image generators and creative programs. The best program for you is going to depend on your project, budget and experience level. You can check out our full reviews and rankings, but these are my three program recommendations.
What’s the best AI image generator for you?
CNET’s Editor’s Choice is Dall-E 3 by OpenAI, the company behind ChatGPT. Dall-E boasts impressive prompt adherence, handling more complex prompts with ease. Its conversational flow — meaning you can follow up easily by typing in the chat window — allows for easy post-generation editing. Dall-E is also one of the more creative programs we’ve tested. You’ll need a ChatGPT Plus account to access the program — $20 per month — and note that OpenAI may use your content for model training. But you can submit a request for it to stop training on your data. (Disclosure: Ziff Davis, CNET’s parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
If you’re looking for a free program, I recommend Leonardo AI. Canva recently acquired it, but you can still use it on its website. Leonardo offers a robust free plan, with more editing tools behind its paywall, and several artistic styles to choose from. It also provides a unique prompt improvement tool, which can take a simple idea and rewrite it, filling in details that make for beautiful end results. Google’s newest AI model, referred to as nano banana, is also a good option if you’re more comfortable using Gemini. You can access nano banana for $20 per month and use the program to edit your existing photos, too.
If you’re really balling on a budget and have no creative experience, check out Canva’s Magic Media generator. It’s extremely intuitive and easy to navigate, which is great for beginners and non-creatives. Canva offers you 50 free credits for image generation (those are lifetime credits — you’ll have to upgrade for more credits on the paid Pro plan after using up your free tier allowance) and automatically opts you out of training its AI on your content.
Once you’ve picked your program, create an account and make your way to the prompt window.
Step 2: Write your image prompt
Your best chance of success starts with a well-written prompt. Every prompt should include the basics of the characters, the setting, other elements and the image’s dimensions. But to really home in on the image you want, add descriptions of the artistic style you’re looking to mimic, the aesthetic, vibe and specific colors.
Leonardo might not understand a prompt that asks for “cottage core coastal grandma,” but it does understand “rustic feel with blues and warm light.”
AI image generators rightfully have limitations around what you can generate. This is to prevent people from abusing these services to create illegal content and potentially dangerous or offensive content. For example, Dall-E 3 will decline to generate images that request public figures by name, like Taylor Swift. Be sure to check out your service’s safety policies before generating.
Prompt engineering — the process of writing the right prompt to get your desired results — is something you get better at with practice. It also varies between each image generator. Canva, for example, tends to do better with simple prompts, whereas Dall-E encourages longer, more detailed prompts. Playing around with your program is the best way to get to know it. You can also check out my complete guide to image prompt engineering for more tips and tricks. Once you’ve got your best prompt, go ahead and generate.
Step 3: Edit your images as needed
Even with the best generator, you’re likely need to edit your images after they’re created. Depending on the service you use, you may have access to editing tools or the ability to request changes. Be wary of over-editing your images. I’ve found that services can only handle so many changes before the image becomes too wonky and ultimately unusable.
Take this image I made in Midjourney as a cautionary tale of what happens when you over-edit. Sometimes it’s best to start anew.
Post-generation editing tools are one area where AI services tend to shine or fall flat. If your service doesn’t offer them (or they don’t work very well), try editing your prompt rather than the picture. In my experience, that’s going to give you better odds of success rather than trying to incrementally edit existing images.
Step 4: Properly credit your images
It’s extremely important to label your images as AI-created content. As AI image services improve, it’s essential that all creators clearly identify the origins of an image, especially to distinguish it from human-created art and photography.
Meta AI watermarks all of its AI-generated images in the bottom-left corner.
Some services automatically add a watermark or other marking indicating it’s AI-generated, but too many don’t, like Dall-E, Adobe Firefly, Midjourney, Canva and Leonardo. If you’re sharing your image online, be sure to include the AI service you used in the caption, credit and/or alt text. You can also add a disclaimer to your posts if you’re sharing on social media. Meta might flag them as containing AI content on Instagram and Facebook, but adding a line to your caption will help ensure there isn’t any confusion.
FAQ and other considerations
As of publication, the use of AI image generators is legal. But there are important legal and ethical considerations you should be aware of when creating AI images. Here are a couple of quick frequently asked questions.
AI image services like Midjourney and Dall-E rely on existing content to help them create pictures. This content, either in a database or on the open web, is typically created by humans. As such, some creators and artists have major concerns about how these programs are designed and trained. That’s led to uproar in the industry and a slew of lawsuits around copyright infringement concerns.
Be sure to take a look at your AI tool’s privacy policy and any disclosures about how the company trains its AI model. Some AI services rely on public databases, like Adobe Stock and Getty Images, where creators sign agreements with the companies to enter their work into those collections for specific uses. Adobe Firefly, for example, says content you create with its AI is commercially safe because it only uses training content from its Adobe Stock public databases and images whose copyrights have expired. It also won’t train on what you create.
As these AI services evolve, the legalities and ethics around them are poised to change as well. If you’re going to be a regular AI creator, make sure you stay up to date with the latest requirements, which we’ll cover in our AI Atlas guide.
