* This blog post is a summary of this video.

Human vs. AI: Graphic Designer Battles Image Generation Bot in Ultimate Creative Showdown

Author: The StudioTime: 2024-02-01 10:55:00

Table of Contents

Introducing the DALL-E Image Generation Challenge

In this blog post, we will recap a fun challenge between MKHBD's graphic designer Tim and an AI image generator called DALL-E. The premise was simple: give Tim and DALL-E the same text prompts to create images from, let them utilize their own unique toolsets, and then compare the results side-by-side. Who would come out on top - human creativity or AI? Read on to find out!

For those unfamiliar, DALL-E is an AI system that can generate realistic images and art from a text description. The capabilities of systems like DALL-E seem to inch closer to human levels every day. We thought it would be an interesting experiment to pit DALL-E head-to-head against an actual human graphic designer.

Over the course of three rounds, Tim and DALL-E were given the same fanciful image prompts focusing on combining disparate objects and styles. While DALL-E relies on advanced AI to churn out images in seconds, Tim had access to Photoshop and stock image libraries to composite images together with no time limit.

What is DALL-E?

DALL-E is an artificial intelligence system created by Anthropic that can generate realistic images and art from a text description. It has been trained on millions of image-text pairs so that it can intelligently translate text prompts into corresponding visuals. The system is named after the iconic surrealist artist Salvador Dali and Pixar's animated robot Wall-E. Much like Dali's unique painting style and Wall-E's unexpected sentience, DALL-E aims to produce novel, thought-provoking images based on textual inputs provided by users.

The Rules of the Human vs. AI Challenge

The rules for the human vs. AI image generation challenge were simple:

  • Tim and DALL-E would be given three identical text prompts to create images from
  • For each prompt, Tim could use Photoshop and stock photos with no time limit, while DALL-E would use its AI capabilities to generate images within seconds
  • The results from both Tim and DALL-E would be compared side-by-side and judged on accuracy to the original prompt as well as overall quality

Round 1: Creating a Deer Covered in Feathers

The first prompt provided to both Tim and DALL-E was: 'A deer covered in feathers'. This whimsical idea combines a common forest animal with an unexpected coat of feathers. Right away, the extremely literal interpretation abilities of AI posed an interesting challenge.

Tim decided to composite images of a deer and a feathery bird together in Photoshop. He paid special attention to differences in lighting, angle and perspective when selecting his base images. Tim also realized matching the feather patterns and colors to the deer's fur would require a careful editing approach.

The Design Process and Decisions

Tim walked through his full creative process using Photoshop to combine images of deer and birds. He demonstrated assessing multiple stock images for consistency of lighting and angle before deciding on his base deer image. Tim also showed the realization that no single bird photo would provide all the right feathers; he would need to carefully select, layer and mask elements from different images. A key challenge was matching the feather colors and patterns to the changes in fur color along the deer's underside. Simply wrapping the whole deer wouldn't sell the concept, so Tim had to push the small details. He also ran into some software crashes during the process, showing the precarious and iterative nature of digital art.

Critiquing and Comparing the Results

When the final images were revealed side-by-side, the DALL-E output was deemed the winner. Though Tim's creation was more literally accurate with identifiable feathers, the AI-generated image just worked visually. The slightly cartoonish fuzziness combined with the deer's puzzled expression made DALL-E's version more successful overall. This round demonstrated the rapid iteration capabilities of AI systems. They cut out much trial and error in translating text to reasonable images. However, determining a clear 'winner' proved complicated - accuracy to the technical prompt or general visual appeal?

Round 2: A Robot Woman Guarding a Wall of Computers

Round 2 presented the prompt: 'A robot woman guarding a wall of computers'. This built upon classic sci-fi tropes with an amusingly specific job assigned to the robot character.

Tim took a cinematic approach, cutting out images of a robot model against an ominous background. He added looming stacks of detailed computer servers. Tim struggled with some software crashes that lost elements like code details on the computers. On the AI side, DALL-E took the textual guarding directive extremely literally - it generated an image of a short robot standing centered in front of, not guarding, computers.

Tim's Creative Vision and Execution

Tim had a grand, dramatic vision for this prompt of a robot defender vigilantly monitoring critical systems. Gathering appropriate sinister, lo-fi sci-fi looking images and carefully clipping detailed elements took hours across multiple days. Software crashes hampered progress at times. In the end though, Tim achieved impressive perspective realism between foreground robot and background computers even if lacking some originally envisioned finishing touches.

Determining a Clear Winner

The consensus winner was DALL-E for better conforming to the specific text prompt. Its humorous literalism of 'guarding' won out over Tim's admirable effort to build a cinematic scene. This demonstrates how in some cases AI interpretation of instructions, while lacking human artistry, can match prompts more exactly. The judges were also intrigued by the AI-generated background code details, presuming those would have required too much effort for a human to invent and placed convincingly. In the end though, both images proved impressively evocative.

Round 3: A Goat Taking a Selfie in Renaissance Style

In the final round, Tim and DALL-E were challenged with: 'A painting in the style of the Mona Lisa of a goat taking a picture with an iPad'. This combined classical high art with a silly modern visual gag.

Tim expertly captured the Mona Lisa's subtle depth and lighting, placing a smirking goat in period attire snapping a quick selfie. The compositing work convincingly carried Renaissance portraiture aesthetics into the digital age. DALL-E also impressed with a creative take incorporating flowing locks, floral backgrounds and an amused goat holding up its hoof.

In the end, judges awarded the win to DALL-E for better conforming to the Mona Lisa stylistic callout, with richer oil painting textures and intricate fine details that would have been beyond Tim's ability to composite together digitally.

Capturing the Old Master Painting Aesthetic

This final prompt combined multiple complex concepts - imitating Renaissance painting styles like depth, lighting and color while inserting an irreverent, modern visual pun with the selfie-taking goat. Tim rose to the occasion, expertly recreating subtle stylistic qualities of masterworks alongside his cheeky composite. The smirking goat in period clothing holding up an iPhone captured the intended mix of classical art homage and internet meme humor.

Declaring the Final Victor

While Tim's work clearly demonstrated his professional skills, DALL-E won out in fully reflecting the fine details described in the complex prompt. The AI-generated version more closely followed the Mona Lisa stylistic callout with richer, flowing locks, ornate floral backgrounds and attractive oils that would not be possible in Tim's composites. The side-by-side comparison here shows how for convoluted or highly aesthetic prompts, AI systems like DALL-E can sometimes outpace even talented human creatives through the sheer depth of their trained imagery associations. That being said, Tim's work required far more complex technical skill and still produced wonderfully clever results.

Key Takeaways and Conclusions

This fun human vs. AI image challenge yielded some intriguing conclusions about the respective strengths of human creativity and computer capabilities:

  • When it comes to strictly literal or technical interpretations of text prompts, AI tends excel through its vast catalogue of image associations

  • On more subjective and nuanced creative goals though, AI still falls short of the artistry, vision and problem-solving skills of humans

Moving forward there is likely room for AI tools and human artists to productively coexist, each focusing on activities best suited to their abilities. Regardless this experiment demonstrated how advances like DALL-E point to a fascinating sci-fi future where technology can emulate and augment human imagination to new heights.


Q: What tools and resources did the human graphic designer use?
A: The graphic designer used programs like Photoshop and Sketchbook along with image libraries and drawing tablets.

Q: How did the AI system generate images so quickly?
A: The DALL-E system uses artificial intelligence and neural networks to analyze text and generate images in seconds.

Q: What criteria determined the winner of each round?
A: The judges chose the image that best matched the specified text prompt for each round.