AI art generators have taken the world by storm in recent years. As someone who enjoys making digital art, I have been eager to test out these new tools to see just how capable they are. In this article, I will provide an in-depth look at the leading AI art generators, evaluating their strengths and weaknesses, to determine just how good they really are.
An Introduction to AI Art Generators
AI art generators are algorithms that can create original images and art from text prompts provided by the user. Some of the most popular generators right now include DALL-E 2, Midjourney, Stable Diffusion, and Craiyon. These tools utilize deep learning techniques like convolutional and generative adversarial neural networks to generate highly realistic and detailed images.
The potential of AI art generators is astounding. With just a few words, they can conjure up photorealistic people, landscapes, objects, and more. However, despite their capabilities, these tools do have some limitations which I will explore throughout this article.
Evaluating Image Quality
One of the most important metrics when judging AI art generators is the visual quality of the images they produce. On this front, the current leading generators are quite impressive in their ability to create convincing illustrations and photos from text alone.
DALL-E 2 and Stable Diffusion stand out for their ability to generate high-resolution, photorealistic images. With the right prompts, they can produce images that are indistinguishable from real photos. The image quality from Midjourney tends to be more stylized and painting-like, while Craiyon produces simpler cartoon-style illustrations.
To evaluate image quality, I tested each tool by giving them the same text prompt: “A still life painting of fruit in a bowl on a table.”
DALL-E 2 produced an image that resembled a real photograph of the scene, accurately capturing the shapes, textures, lighting, and colors of the fruit and bowl.
Stable Diffusion also generated a photorealistic image, but some of the fruit shapes were warped.
Midjourney’s rendition had a much more impressionistic, painterly style. The lighting and colors were appealing but the shapes were more ambiguous.
Craiyon’s output was the most basic, with simple 2D fruit and shapes. The image lacked realism but conveyed the gist of the prompt.
So in terms of image quality, DALL-E 2 and Stable Diffusion are superior, but the stylized outputs from Midjourney have their own aesthetic appeal.
Evaluating Creative Abilities
While AI generators can produce realistic images, how creative can they really be? To evaluate creativity, I tested the tools by giving more imaginative and open-ended prompts.
For the prompt “An astronaut riding a unicorn in a neon city,” Midjourney generated the most artistic, visually striking image. The unicorn was beautifully animated, and thefuturistic cityscape had incredible detail.
DALL-E 2’s output was photorealistic but a bit visually dull.
Stable Diffusion struggled with this abstract concept, generating a nonsensical image.
Craiyon’s attempt was simplistic but conveyed the general idea.
Testing with other prompts revealed Midjourney to have superior creative abilities for original concepts, pop culture, and sci-fi/fantasy themes. However, its images are more stylized. DALL-E 2 does well at creative prompts if they have some grounding in reality.
So for creativity, Midjourney seems to have an edge, but the other tools also have strengths depending on the type of prompt.
Evaluating Versatility
The versatility of AI image generators – their ability to successfully handle a wide range of concepts and art styles – is another important consideration.
To test versatility, I gave the prompt: “Impressionist landscape painting of a rainy blue city with tall buildings”
Midjourney again excelled, generating an evocative landscape accurately capturing the Impressionist style.
DALL-E 2 also did well, blending photorealistic skyscrapers with Impressionist brushwork.
Stable Diffusion struggled with the ambiguous concept, producing a nondescript city scene.
Craiyon managed to depict the key elements of rain and buildings, but in its simple cartoon style.
In my testing, Midjourney demonstrated the most versatility in terms of handling diverse subjects, styles, and artistic movements. DALL-E 2 was also adaptable, but sometimes defaulted to a more photorealistic style.
Evaluating Conceptual Abilities
Another key test is how well AI art generators can understand broader concepts, contexts, and deeper meaning when creating images.
To evaluate conceptual abilities, I gave the prompt: “A poster that promotes sustainability using the text Reduce, Reuse, Recycle.”
DALL-E 2 produced the best image for this prompt, deftly integrating the three R’s in a visually impactful poster composition.
Midjourney and Stable Diffusion both struggled, generating bizarre images that didn’t really capture the concept.
Craiyon was able to depict the three words but lacked any meaningful poster composition.
Additional testing revealed DALL-E 2’s superior ability to interpret conceptual prompts and incorporate contextual elements. The other generators sometimes captured the literal words but failed to convey deeper meaning.
Comparing Text-to-Image Capabilities
One advantage of AI art generators over traditional digital art tools is their ability to generate images directly from text prompts. But some tools excel more at this text-to-image synthesis than others.
DALL-E 2 sets the standard for coherent text-to-image generation. It consistently produces realistic images that accurately match the prompt’s description. The text understanding abilities of Stable Diffusion are nearly on par with DALL-E 2 as well.
Midjourney has outstanding creative abilities, but the connection between text prompt and final image is not always aligned. Craiyon’s simplicity limits how well it can render detailed text concepts.
So for precise text-to-image generation, DALL-E 2 and Stable Diffusion are preferable, but Midjourney takes the lead for artistic interpretation.
Comparing Accessibility and Ease of Use
The user experience is also an important consideration when evaluating AI art generators. Factors like accessibility, cost, and ease of use determine how feasible they are to integrate into creative workflows.
DALL-E 2 remains the most limited in accessibility as it is still in closed beta. Getting access requires joining a waitlist.
Midjourney requires a Discord membership and has a steep learning curve for mastering its text prompt commands.
Stable Diffusion is open source but requires setting up locally and experimenting with different UI options.
Craiyon is the most accessible, available to use via web browser for free without any waitlist. Its text-to-image feature is quick and intuitive.
In terms of ease of use, Craiyon wins for its no barrier simplicity. Midjourney and Stable Diffusion have more capabilities but also more friction in the user experience. DALL-E 2 seems likely to offer an elegant interface whenever it becomes widely available.
Evaluating Originality of Output
A major concern with AI art generators is a lack of originality in their output. Some are more prone to repeating common image tropes and cliches.
In my testing, images from Midjourney displayed the most originality and uniqueness across various prompts. Even when given the same prompt multiple times, it consistently generates novel interpretations.
DALL-E 2 does decently well at originality for photorealistic prompts, but less so for abstract concepts. Stable Diffusion tends to stick closely to what’s present in its training data, resulting in commonplace images.
Craiyon’s simplicity means it recycles its limited visual vocabulary often.
So for true originality, Midjourney shows the most promise, with DALL-E 2 also doing well for grounded prompts. But the other tools struggle with repetitiveness at times.
Compared Generative Capabilities
Some AI art generators allow users to iterate on images, editing and evolving the outputs through multiple generations. This evaluative capability is powerful for refining images towards a desired result.
Midjourney stands out with its robust generative features. You can resubmit prior images to the bot and receive variants exploring new directions. The tool also lets you edit images by adding or removing specific elements.
DALL-E 2 and Stable Diffusion currently lack extensive generative options, although these are likely in development.
Craiyon has no generative capabilities. Each image is a one-off creation.
So Midjourney offers the most generative flexibility, allowing for deeper creative exploration of concepts. The other tools mostly produce single static outputs.
Summary: AI Art Generators Have Come Far, but Have Limitations
In reviewing the leading AI art generators, they have clearly come a long way in their capabilities. However, each tool still has weaknesses that limit how suitable they are for creative professionals.
DALL-E 2 excels at photorealism and interpreting text concepts, but falls short on abstract creativity.
Midjourney is unparalleled on originality and imagination, but lacks precision.
Stable Diffusion generates decent quality images but struggles with creativity and versatility.
Craiyon is accessible and easy to use, but very basic in its outputs.
AI art has incredible potential, but for now human artists still reign supreme in their versatility, originality, and creative dexterity. These tools are impressive and can augment human creativity, but have not yet replicated it.
The rapid pace of development in AI means art generators will only grow more advanced in the near future. For now they present artists with new opportunities for experimentation, inspiration, and expanding their practices. But a full replacement for human artistic abilities remains elusive. Evaluating their strengths and weaknesses helps illustrate the accomplishments and limits of current AI art technology on its journey towards matching human creativity.
