A 4-year-old blonde girl holding a pink umbrella
While there are many different AI image generators available, DELL-E 2 is one of the most comprehensive and user-friendly options. With access to it now, we are able to compare the different AI image generators on one of our favorite prompts. The results are often quite striking and provide valuable insight into the capabilities of each generator.
One of the most difficult challenges for artificial intelligence is generating faces and geometry that look realistic. This is because the human eye is incredibly good at spotting even the smallest irregularities. For example, consider the spokes of an umbrella. Each one is slightly different in shape and size, but our brains automatically correct for these differences when we see them. As a result, it can be very difficult to create geometry that looks completely realistic. However, some progress has been made in this area in recent years. For example, new algorithms have been developed that can generate realistic 3D faces with accurate lighting and shading. These advances suggest that it may eventually be possible to create AI-generated faces and geometry that are indistinguishable from the real thing.
Our test prompt is “a 4-year-old blonde girl holding a pink umbrella“.
DALL-E 2
DALL-E 2 is the clear winner in this comparison. It creates exactly what this prompt, inspired by Steve’s daughters’ love for pink umbrellas, asked it to create. Faces can still be improved, but generally these images are quite impressive already. Apparently OpenAI’s new “GLIDE” (September 2022) model will be even better.
Stable Diffusion (DreamStudio)
Stable Diffusion has a knack for creating wonky faces and definitely has no concept of differentiating how many people a prompt is describing. The geometry of umbrellas is already much better than with Midjourney and NightCafe.
Midjourney
Midjourney creates good quality artwork but it struggles to create realistic faces that are not scary. It also has no more than a very artistic understanding of how to create the geometry of real-word objects like umbrellas.
NightCafe
Modified prompt due to “prohibited words“: “a young year old blonde girl holding a pink umbrella“
NightCafe is basically useless for this kind of prompt.
There are many more AI image generators out there, but these are currently the most popular and prominent ones.
Want to learn more about the different image generation models?
This video goes into detail about different image generation models, such as GANs, VAEs, flow-based models, and diffusion models. It also looks at OpenAI’s new GLIDE - their next iteration on diffusion models.
