The AI image generation market is super hot right now. Social media, marketing, game design, prototyping – there’s a constant need for fresh visuals everywhere. The endless stream of creativity that you can produce using tools like Stable Diffusion and Midjourney is promising, exciting, and thrilling. Well, maybe less so for the actual artists that draw for a living. But if you’re a creative professional, they’re real breakthroughs. The creation of high-quality visuals is now more accessible than ever for everyone.
We’ve seen an explosion of tools hitting the scene, but which one should we pick? It’s not an easy answer, considering the significant competition. Fortunately, we’re here to help. We’ll showcase the four best tools that you can use for image creation. We will also provide examples and sample prompts to show you the capabilities of each AI image-generation tool.
Dall-E: Easy to use, built into ChatGPT
Developed by OpenAI, Dall-E is a cutting-edge AI program designed to generate digital images from textual descriptions. You can use it straight from ChatGPT, although at this moment, you will need GPT4 to create images. It’s particularly noted for its ability to blend concepts in unexpected ways, producing images that are not unique but sometimes startlingly inventive.
Dall-E Pros:
- Quite creative: DALL-E can generate highly original and creative images from textual descriptions.
- Easy to use: Just type the description of the image in GPT4 to start generating images. Dall-E has the lowest learning curve.
- Versatile: Dall-E can generate in many different styles and, in general, has a wide range of applications thanks to its extensive image tokens database.
Dall-E Cons:
- Very restrictive: Since it’s close-source, it has many “safety” features that won’t allow you to generate any images considered adult or copyrighted (i.e., you can’t generate Super Mario) and skews generations towards “diversity.” Dall-E is very strict in this regard.
- Inconsistent in style: It currently features a single model, and even with careful prompting, it’s not easy to achieve consistent style by generating a series of images.
- Limits on the number of generations: no more than 20-40 images per 1 hour.
- Paid software: Costs $20/month to access GPT4 and Dall-E.
Dall-E showcase: Sample prompts and images you can generate with Dall-E:
“A futuristic cityscape at dusk, with flying cars zooming between skyscrapers that are adorned with glowing neon lights, reflecting off the surfaces of buildings and the water below. The scene combines elements of advanced technology with a hint of cyberpunk aesthetic, showcasing a world where technology and urban life merge seamlessly.”
“A whimsical illustration of a woman surrounded by a magical forest. The woman is depicted with an aura of kindness and joy, interacting gently with fantastical creatures around her. She wears a dress made of leaves and flowers, blending seamlessly into the enchanting woodland environment. This scene captures the essence of a fairy tale, highlighting a connection between humans and nature in a creative and positive light.”
“A hyper-realistic image of a fashionable young Asian woman walking through an urban landscape. She radiates confidence and style, showcasing the latest urban fashion trends. Her ensemble features a harmonious blend of modern chic and streetwear, complete with a tailored jacket, graphic tee, and sneakers. The bustling city backdrop includes skyscrapers, neon lights, and street art, providing a vivid and dynamic setting. This image captures the essence of contemporary urban fashion, emphasizing the woman’s individuality and the vibrant energy of the city.”
Adobe Firefly: Standalone Image gen software built into Photoshop
Adobe Firefly, created by the software company Adobe, of Photoshop and PDF Reader fame, is designed to integrate with Adobe’s suite of creative products. This AI-powered tool focuses on generating assets and elements such as textures, patterns, and specific types of images that can be directly used in projects within Adobe’s ecosystem. So, if you’re already using Photoshop, you can integrate your generations seamlessly into your work.
Another thing to note is that it’s trained on its database of images. Unlike any other AI model that has been trained on vast datasets “sourced” from the Internet (or grabbed without asking), Adobe has been very clear about its approach to training Firefly’s models. Adobe trained its AI generation model on images, illustrations, and other assets from a database of Adobe Stock they own. This, in general, addresses concerns related to copyright and ethical use of digital content in image generation, as all the image owners consented to grant the rights to Adobe.
Unfortunately, Adobe Firefly is also paid software, and it is rather expensive. It uses credits as a form of currency generation, and you’re paying several cents up to 1 dollar for each generation. The price makes it a less popular option, but still something to consider, especially if you’re a Photoshop user.
Adobe Firefly Pros:
- Full integration with Adobe Photoshop: directly generate, manipulate, and inpaint images in Adobe Photoshop. Very useful for pro graphic designers.
- Web interface available: Feature an easy-to-use web interface that allows you to tweak styles and input additional parameters. It offers more control than Dall-E and Midjourney but less than a Stable Diffusion.
- High-resolution outputs: Base images are generated at 2048×2048 resolution, compared to 1024×1024 in the case of Dall-E.
Adobe Firefly Cons:
- Very restrictive: Similar to Dall-E and Midjourney, it doesn’t allow for many keywords.
- Limited training data: Since Firefly’s training data is confined to Adobe-owned content, there might be a ceiling to how novel or varied the generated outputs can be, especially in comparison to models trained on broader datasets. See the generation of World War 2 soldiers below, which shows that, despite the lack of training in photography from that period, the Firefly model outputs historical nonsense.
- Most expensive: You’re paying for each generation. If you intend to generate a lot of images, it’s not a viable option. Subscription cost is probably the highest out of all the tools that are now available on the market.
Adobe Firefly showcase: Sample prompts and images generated with the web interface
“dog in a sweater, colorful, happy”
“cyberpunk woman samurai with sword and bird in the smoke, anime style”
“photography of a world war 2 american soldier, tired, black and white, historical”
Midjourney: The first commercial image generation app
Midjourney was pretty much the first commercial image generation software that offered quality outputs. To this day, it is still one of the best choices for the task. It’s simple to use, and the users can interact with Midjourney through a variety of interfaces, including a widespread integration with Discord. A huge dataset and correct prompt reading make for some gorgeous artwork that can be produced.
Unfortunately, Midjourney is far from being perfect. Using a system of credits can be expensive, as you’re paying as you go; each generation costs you several cents, up to less than a dollar. This payment model is similar to Adobe Firefly. Moreover, Midjourney is heavily censored, with a very long blacklist of keywords. It throttles the creativity and makes it less usable than it should be. Still, it’s easy to use, and it’s certainly suitable for beginners – but be aware that it can be pretty costly, even when used professionally.
Midjourney Pros:
- Easy to use: Even concise prompts can lead to superb designs. Doesn’t require extensive prompting for effective results.
- Available through Discord: It’s an odd design choice, but it makes it simple to use.
- Huge dataset for a generation: The vast dataset makes for some stunning creation and accurate, prompt interpretation.
Midjourney Cons:
- Prone to censorship: Extensive content censorship limits creative freedom. Has an infamously long keyword blacklist, to the point you can’t generate a “slim woman.”
- Most expensive: A credit-based payment system is costly. If you intend to generate a large amount of images, the costs will accumulate quickly, leading it to not being worthy for non-professional uses.
Midjourney showcase: Sample prompts and images generate with the web interface
“A page from the anime “Sailor Moon”, Sailor Moon is playing with her cat. Sailor Moon has blonde hair and blue eyes. In one panel she is very happy and smiling. In another panel there is a question mark above her head, with pink roses around them. At the top it says “sailor””
“A lovely little Chinese boy, He wears light-colored Hanfu, Riding a water buffalo, IP by pop mart, In the field, There is a river in the distance, grass, there are distant mountains, Flying bird, Lush greenery, Disney Pixar style character, Watery eyes, Bright color, raytracing, octane rendering, Wide Angle, vision, clay materials, animated lighting, 3D, C4D, OC rendering, Blender, IP, best quality, ultra-high definition, 8K”
“minimalist illustration of a tennis player with white shoes and pink shorts holding a racket on a green court, in the style of Gemma Correll, isometric style, flat design, sunshine, sports atmosphere, minimalism, behance, dribble, simple background, closeup shot”
Stable Diffusion: Open-source generation on your PC
Unlike all these previous commercial image generation suites, Stable Diffusion runs on your PC. The way it works is that you download the models you want to use (usually around 4-8 GB size each) and then generate images using your own GPU or graphics card.
Also, unlike its commercial counterparts, it offers creative freedom without censorship or content restrictions. Running Stable Diffusion locally on your PC is a bit more complicated. Still, once you set it up, it provides privacy, customization, and independence from cloud-based models, making it the best choice for both personal and professional use. Typically, you will use Stable Diffusion with a GUI (or Graphical User Interface), like Automatic1111 or Comfy UI.
As mentioned, Stable Diffusion software and a myriad of its models are free, but you will need a PC with at least 6 to 8 GB VRAM graphics card – the better your graphics card, the faster your images will generate. On a typical gaming PC, generating a single image takes from a dozen seconds up to a few minutes, depending on the resolution, steps, and other settings. Currently, Stable Diffusion XL is the latest iteration, but the new SD3 promises even more awesome outputs.
Stable Diffusion interface is a bit more complex, doesn’t run from a website, and requires installation on your PC, but neither Dall-E nor Midjourney can match the quality, creativity, and style adjustments that you can get from SD.
“a glass Apple sculpture, concealed inside is a picture of a beautiful Japanese landscape inside of it with pink trees and falling leaves, in the dark, detailed image, 8k, 8k high quality detailed art, the moon, perfect sphere, amazing wallpaper, digital painting highly detailed, 8 k hd detailed oil painting, beautiful art uhd 4 k, focus on glass apple, close up, bokeh, unfocused counter top background “
“1flower in a concrete jungle with plants and flowers at night, cinematic, wlop, concept art, digital painting, trending on artstation, highly detailed, epic composition, official media, 8 k uhd<lora:MJ52:0.5> <lora:add-detail-xl:1>”
“perfect face,(Bruce Lee:1.5) martial art masterpiece inspired by Crouching Tiger, Hidden Dragon,in the background a Chinese dragon, stylized art”
“1 anime wolf man, male, furry, kawaii, vintage anime style, soft pastel color, school uniform, soft color, grey, detailed, HD, 8k, handsome body”
“(hdr, dark shot:1.2), closeup portrait photo of beautiful woman, 25 years old, smiling, makeup, dramatic eyeliner, vibrant silver hair, sexy, simple black background, 8k uhd, high quality, dramatic, cinematic, vibrant colors, high contrast, (natural skin texture, hyperrealism, soft light, sharp)”
” city street, (((older female))), braid, cargo pants, tank top, [handbag], medium brown hair, happy”
Stable Diffusion Pros:
- Open-Source and Private: It’s transparent and available for everyone to download. Running it on your own PC also offers you a privacy, as unlike in Midjourney or Adobe Firefly, your outputs are stored on your disk only. You have full control over your work and data security.
- Fantastic capabilities: Stable Diffusion imposes no content restrictions and has thousands of models available for many specialized tasks. No other AI generation software offer such quality and artistic freedom.
- Free to use: Unlike many AI image generators that require subscriptions or per-image fees, Stable Diffusion is completely free. You run it on your PC, instead of in paid cloud services and you can utilize it as much as you want.
Stable Diffusion Cons:
- Requires Technical Setup: Unlike cloud-based, it’s not ready out of the box. Stable Diffusion demands a bit of technical know-how to get started. Setting it up on your computer involves downloading software and possibly troubleshooting compatibility issues. But the setup is ultimately 15 minutes + additional time to sift through models and extensions to use, so it’s not that complex.
- Hardware Requirements: To run Stable Diffusion effectively, you need a relatively powerful computer, especially one with a good GPU. 4 to 6 GB VRAM is a minimum, and the more powerful your graphics card is, the faster your will be able to generate outputs.
- Overwhelming Options: The huge array of models and customization options is a pro, but it can also be a con for users looking for a simple, straightforward experience. It’s definitely more complex than Midjourney or ChatGPT (Dall-E) but it shouldn’t deter you, even if you’re just starting.
There are a few other AI-gen tools, but so far, none can compete with the software presented below. In the end, while the competition remains fierce and each tool has its merits, Stable Diffusion is the clear winner, especially for more creative use like digital artwork, prototyping, making comics, or just fun. Check out Stable Diffusion articles and guides if you’d like to learn more about it and start creating cool stuff!
