I have created a lot of content using generative AI platforms. One of the first things I do every morning is open Ideogram and create something random, animate it in Runway and create a voiceover with Hedra. I have tested multiple AI models, particularly image models and they each have something unique to offer.
AI image generation has come a long way in a few short years. The first version of Midjourney came out in February 2022. It could barely create a realistic-looking human. Now, almost three years on its output is indistinguishable from reality.
Recent updates
Update Nov 8: We've done a face-off between Apple's new Image Playground vs Google Pixel Studio to see which AI image generator wins.
In a rapidly evolving market, a good model isn’t enough anymore though. What makes a platform stand out is the additional features including tools like Ideogram’s Canvas, upscaling in Freepik or Midjourney’s Editor which also includes editing external images.
Working out which model to use, especially when you are parting with money to get the best capabilities, can be a challenge. Hopefully, this guide will help. As well as simply using the image generation platforms, I also put them to the test with a range of prompt styles, image types and criteria to see not only how they perform, but also how they compare to one another.
I've run realism tests, which image generator animated best with Runway, which creates the best text and which produces the most lifelike people. All of these helped me come up with this guide to the best AI image generators available today.
How we test AI image generators
Why you can trust Tom's Guide Our writers and editors spend hours analyzing and reviewing products, services, and apps to help find what's best for you. Find out more about how we test, analyze, and rate.
The most simple test of any tool is to use it — a lot. I create images as part of my everyday work in all of the big AI image generators, especially the ones I’ve included here. I then run comparison tests, and prompt tests, and try to see what they can’t do.
I play with different tools and features not part of the core model and see how the output interacts with other tools such as AI-generated video or lip-synching. Each model has its own unique prompting quirks as well, and I’ve tried to find the best of those.
For this guide, I’ve focused on platforms rather than underlying models. Some models, like Flux and Stable Diffusion, are widely available on services like FAL, NightCafe, Leonardo and Freepik. Others are single-platform models. However, that is changing with the rise in APIs. For example, you can get Imagen 3, DALL-E 3 and Ideogram on NightCafe now.
For each of the generators below I came up with five prompts to put them to the test. You can see the resulting images in a gallery for each section. For Leonardo, these were made with Phoenix, NightCafe images were made using Flux 1.1 Pro and Freepik images were generated using its own Mystic model.
Ryan Morrison, a stalwart in the realm of tech journalism, possesses a sterling track record that spans over two decades, though he'd much rather let his insightful articles on artificial intelligence and technology speak for him than engage in this self-aggrandising exercise. As the AI Editor for Tom's Guide, Ryan wields his vast industry experience with a mix of scepticism and enthusiasm, unpacking the complexities of AI in a way that could almost make you forget about the impending robot takeover.
Best AI image generator overall
While Midjourney doesn’t have the range of models or video capabilities of the big platforms like FAL, NightCafe and Leonardo, it is tried and tested. It is the power user AI image generator with high levels of realism and an impressive degree of control over the output.
It took me a while to get to grips with the huge number of parameters and features available through Midjourney, especially while it was still inside Discord, but once you’ve mastered the fundamentals it is a remarkable application.
Some of the more talented users can get it to create pictures that look like they’re straight from a phone camera. Midjourney was one of the first to crack the finger problem and has consistently real-looking people. With version 6.1 the company added new skin textures to further improve human realism.
Midjourney is somewhat controversial for its refusal to discuss the source of its training data. Many suspect much of it comes from scraping any publicly available images it could find regardless of whether it had permission from the image creators.
What makes Midjourney stand out for me though is the level of control you have over every aspect of the generation. You can use parameter commands to reference the style or a character within another image or use others to completely change how an image appears.
Best for Features
Leonardo started life as a well-done wrapper for the variety of Stable Diffusion models but it has evolved into so much more than that with the addition of the powerful new Phoenix model. It still has the SD models but offers them with custom styles and fine-tuned versions of models.
Leonardo has also joined Affinity and become part of the Canva family. The foundation models built by the Australian image startup will likely be integrated into Canva products and the sale will give Leonardo the resources to expand and build more impressive generative AI features.
It is already up there with the best. Leonardo's ability to generate photorealistic images, thanks to the fine-tuned PhotoReal model, is nearly up to Midjourney standards and it can create a range of styles with the Elements feature.
These elements are a fine-tuned model that can be applied before generation and steer the image to be created with a particular look such as a sketch or sculpture. You also have the ability set a style such as cinematic, food or long exposure.
What really makes Leonardo standout for me is the combination of an easy to use UI, mixed with an incredible amount of control. You can add reference images and set how the AI should use them, control the size and layout, even add a transparent background.
Most of these features are available in other platforms, but Leonardo has them all as well as a range of other features such as image upscaling, live image generation and one of the most creative tools — the ability to draw a sketch and have the AI turn it into a full image.
Best for text on images
Ideogram is one of my favorite AI image generators for my own personal use. Version 2 gave us a new set of features that put it on par with Midjourney, as well as some not available elsewhere such as color pallets and a 'design' style.
It follows a prompt really well and can add text like no other model. I’ve been able to generate full movie posters, flyers and greeting cards with accurate text.
With the new Canvas feature you can edit and manipulate generated images using extended text prompts. This is great for correcting text or graphic design.
Accessed through a wonderfully simple prompt box, with the option to automatically enhance your prompt to get a better image, it is both easy to use and powerful.
While it is best for adding text on to images, it also has a bit of Midjourney flare to the style of the works it generates. You can turn off magic prompt and create more artistically simple images, or even add customized style tags.
Ideogram's Magic Prompt is a compelling feature. If its turned on then a large language model will analyze your prompt and re-write to be much more descriptive to get closer to your vision.
You can see your original prompt as well as the magic prompt for any image, adapt it or use it to create a new picture. You can also use any generated image as a source for a new image.
Best for collaboration
Google Gemini has had a rocky road when it comes to image generation. It uses the Google DeepMind developed Imagen family of AI image models to create pictures with the Gemini chatbot acting as a go-between, enhancing prompts and making changes.
This led to some issues where it was generated unexpected images of people. Google blocked Gemini from making pictures of people — real or fake — for a while and limited the rollout of image creation globally until it was fixed.
With the launch of Imagen 3 earlier this year Gemini was given its image creation ability back and, while it can only make square format pictures, its collaboration capabilities are what make it stand out.
Once you get the first round of pictures from a prompt you can then ask Gemini to change any single aspect. In one test I was able to ask it to generate a grand piano, then add a cat at the keys and finally turn it into a full band with a giraffe on bass and dog singer.
Best for interaction
DALL-E 3 is only available in ChatGPT to those with a Plus account. There are a few ways to use DALL-E within ChatGPT. You can just access it through the main interface, through the DALL-E GPT custom chatbot, or by tagging DALL-E in the main chat.
The original DALL-E was one of the first high profile commercial generative AI image tools. Initially available as an API or through a dedicated DALL-E page, OpenAI has since bundled it with its chatbot. This is also its primary selling point — the ability to talk through an image.
Everything is based on text prompts and it uses completely natural language for generation. For example you can tell it to generate a picture of a cat, then follow up with asking it to add a hat.
With the most recent update you can now click on the image itself and make edits within the picture. This is done by drawing over the part you want to change and telling ChatGPT how to change it — once again relying on the conversational nature of the editor.
I don’t think DALL-E is the best AI image generator but it is a good all rounder. It can do text, it can create photorealistic images (with a slight uncanny valley) and produce artistic work but the ability to reason and rationalize over the image with text is the big benefit.
It is in urgent need of an AI image update. Technically GPT-4o can create images on its own without the need for a third model (the way Gemini uses Imagen 3 or ChatGPT currently uses DALL-E 3) but it isn't currently allowed to do so because of safety concerns.
Best for ethical training
Adobe Firefly has some impressive tools including prompt suggestions, deep customizations for the image generation and a training dataset almost exclusively trained on Adobe Stock pictures.
This last point means it has a more ethical training set than most image generators on the market, even prompting Adobe to offer financial indemnity from copyright claims against images generated using Firefly.
Firefly was given a major upgrade recently with the arrival of the Firefly 3 model. With this came impressive degrees of photorealism compared to the previous model and even better artistic skills — which were already among the best.
Adobe offers a number of generative AI features including vector generation, template creation and generative fill in Photoshop, all powered by the Firefly model.
One of the best features of Firefly is also one of the most recent. It is called Structural Reference and lets you carry the layout of one image over to another.
There is even a video model coming soon.
Best for community
NightCafe was one of my first introductions to the world of generative AI. It was initially built on some custom models and later expanding to use Stable Diffusion-based models that have since been fine-tuned and customized.
As well as offering access to a wide variety of models, including Flux, Google's Imagen 3, Stable Diffusion, OpenAI's DALL-E 3 and now Ideogram, NightCafe makes it relatively easy to fine-tune or train your own models and lets easily manage and share your creations.
It even allows you to animate the individual photos using Stable Video Diffusion and upscale images using a range of upscaler models. You can even animate with Runway Gen-3 Turbo and Luma Labs Dream Machine within the single platform.
Its community is the real secret sauce behind why I keep going back to NightCafe again and again, regardless of which new models or services launch. There are regular competitions, chatrooms and creator jams for new ideas.
The company works to support that community, even engaging in partnerships with third party companies to offer prizes and free credits.
The credit system is also particularly helpful. Other services require a fixed monthly payment or have costly credit packs but NightCafe allows you to buy what you need.
It isn't perfect. The UI can be busy and the initial user journey can be daunting, but once you've engaged with it a while you'll find yourself coming back again and again.
Best for Innovation
Freepik is an interesting platform. Originally know for its library of images, it has very quickly become a major force in the artificial intelligence space. Some of this was driven by its acquisition of AI-upscaler Magnific earlier this year.
The AI generation part of the platform is incredibly easy to use and intuitive. It also includes a prompt enhancer that uses a large language model to improve your simplistic prompts.
You can add filters and elements to the image generation such as custom styles, colors and lighting. You can also specify the camera type if you are taking a photo. When exporting you have an easy upscale button that happens automatically before offering the download.
The feature set goes beyond simple image generation. One of these is sketch enhancement using AI. You can doodle on the screen and the AI will turn your doodle into an artwork or photograph.
Outside of generation, editing and sketching you can also upscale images from outside of Freepik, expand canvas in a similar way to Photoshop and use AI to retouch existing images.
Freepik has become a good all-round AI powerhouse, a way to play with image creation, including offering the latest models. It has all versions of the Flux family of models including Flux 1.1 Pro, the new rapid and realistic version.
The big stand out though is Mystic, built by the Magnific team. This is a fine-tuned and customized version of Flux that is perfect for photorealistic depictions of people.
Final thoughts
Generating AI images is a fun process. You get to see an idea you've had in your head almost immediately and with new features like Canvas you can edit and make changes to that image to fit your exact concept.
I've used Ideogram to create posters to remind the children to turn off lights, artwork for app icons and even entire app designs. I've used Midjourney to create photos worthy of printing and Adobe Firefly regularly contributes to cover images for my stories.
I hope this guide has given you some insight into the best tools for creating AI art. If you've tried something I've not included send me an email, find me on X or leave a comment and I'll give it a review.
More from Tom's Guide
- OpenAI's GPT Store is now live with over 3 million custom chatbots to try
- OpenAI custom chatbots — this is how they work and what they can do
- Forget ChatGPT — OpenAI now lets you build your own chatbot for any task
