I Gave the Same Prompts to ChatGPT and Midjourney… Who Won?

Key Concepts

• Image Generation: Comparing capabilities of different AI models for creating images.
• Prompt Adherence: How well the AI model follows the instructions given in the prompt.
• Aesthetics: The visual appeal and artistic quality of the generated images.
• Text Generation: The ability of the AI model to accurately generate text within images.
• Coherency: The logical consistency and realism of the generated images.
• Character Consistency: The ability to maintain the same character design across multiple images.
• Midjourney V7: The latest version of the Midjourney AI image generation model.
• ChatGPT (GPT-4o): The AI model from OpenAI, integrated with image generation capabilities.
• Personalization: Training the AI model on user preferences to generate more customized images.
• Upscaling: Improving the resolution and quality of a generated image.

Portrait Photography

• Basic Portraits: Midjourney produces more authentic and lifelike portraits compared to ChatGPT, which tends to generate images that appear more posed and edited.
  • Example: Water droplets and depth of field are more realistic in Midjourney's output.
• Diverse Portrait Prompts: Midjourney generally outperforms ChatGPT in terms of aesthetics, but ChatGPT's images often suffer from oversaturation and a slightly too warm yellow hue.
  • Example: White balance adjustments can significantly improve ChatGPT's images.
• Hands: ChatGPT excels at generating realistic and anatomically correct hands, while Midjourney still struggles with this aspect.
  • Example: In images featuring hands holding objects or performing actions, ChatGPT's output is more accurate.
• Rock Paper Scissors Test: ChatGPT demonstrates superior prompt understanding by accurately depicting the game of rock paper scissors, while Midjourney fails to grasp the concept.

HubSpot AI Website Generator

• HubSpot offers a free AI website generator that allows users to create custom websites by answering a series of prompts about their business.
• The AI generator creates a site with no coding required, and users can tailor it with drag-and-drop editing.
• HubSpot provides hosting and mobile-friendliness, and the AI-built website integrates with other HubSpot tools for SEO, email marketing, and lead generation.

Cinematic and Action Shots

• Cinematic Look: Midjourney generally leads in aesthetics for cinematic shots, but it can struggle with prompt adherence and coherency.
  • Example: A clown image generated by Midjourney contains a gigantic iron, and the body in the background is morphed.
• Action Shots: Midjourney's style is appealing for action shots, but it can have issues with faces, especially in group settings.
  • Example: Using a creative upscale in Midjourney can sometimes fix face issues.
• Candid Look: Both Midjourney and ChatGPT perform well in generating unpolished and candid-looking images.

Complex Prompt Adherence

• ChatGPT demonstrates superior prompt adherence, accurately capturing specific details and complexities in prompts.
  • Example: ChatGPT accurately generates an image of a woman with specific clothing, hair, jewelry, and tattoos, while Midjourney misses multiple parts of the prompt.
• Midjourney struggles with text generation, producing garbled and nonsensical text even in simple prompts.
  • Example: ChatGPT accurately generates a fake Wikipedia page with a description written by the AI, while Midjourney fails to produce a coherent image.

Complex Prompts and Actions

• Object Placement: ChatGPT accurately places objects in the scene according to the prompt, while Midjourney sometimes struggles with this aspect.
  • Example: ChatGPT correctly depicts a dog sitting on top of a cube, while Midjourney fails to do so.
• Chess Board: ChatGPT generates a chess board with alternating sapphire and marble tiles and accurately depicts the chess pieces according to the prompt, while Midjourney's output is less accurate.
• Upside-Down Faces: Midjourney often struggles with actions or poses where the face is upside down, resulting in distorted and funny-looking images.

Crowds and Censorship

• Crowds: Midjourney struggles with generating realistic faces in crowds, often resulting in morphed and distorted faces. ChatGPT performs better in this area.
• Censorship: Midjourney is more relaxed with IP, especially when it comes to big names like Disney or Pixar, while ChatGPT censors some of these. Both models have limits on what public figures can be doing.

Likeness and Anime Styles

• Likeness: Midjourney struggles with likeness when venturing outside of ultra-famous people, while ChatGPT can generate accurate likenesses of a wider range of individuals.
• Anime Styles: Both models can generate images in the style of Studio Ghibli, but ChatGPT struggles with other anime styles when asked for them by artist name. It works better if asked by movie title.

Other Styles and Surreal Prompts

• Other Styles: Both models can generate images in various styles, such as tilt-shift, faux-naïf, and Pixar style.
• Character Consistency: ChatGPT is better at generating multiple scenes with the same character, as Midjourney currently lacks character reference tools.
• Surreal Prompts: Both models perform well with surreal prompts, but Midjourney sometimes misses specific details.
• One-Word Prompts: Midjourney leads in default creativity and aesthetics, especially when generating abstract or weird images.

Conclusion

Midjourney generally produces more aesthetically pleasing images, but ChatGPT excels in prompt adherence, text generation, anatomy, coherency, multiple faces, and consistent characters. Midjourney is faster, but ChatGPT is more accurate, especially when specific details are required. Both models have additional features, such as editors and draft modes, but ChatGPT's slower generation speed can be a limiting factor.