ChatGPT's Image Capabilities vs. Midjourney V7: Which AI Wins?

AI image generation tools keep getting better. Recently, ChatGPT got huge new image features, and Midjourney released its highly anticipated Version 7. Both can create amazing pictures, but how do they compare when given the exact same tasks? We tested them head-to-head across many styles and challenges to see what each does best.

How the Test Worked

We used Midjourney V7 with default settings and ChatGPT's image tool (part of GPT-4o). ChatGPT's aspect ratios are limited, so we used 3:2 or 2:3 for those. Midjourney V7 required ranking 200 images first for personalization, but this was turned off for consistency across models. Best results from a grid of four were chosen for Midjourney, while ChatGPT typically generated two at a time.

Comparing Portraits

For simple portrait prompts, both models followed instructions well. However, Midjourney's results looked much more real and lifelike. ChatGPT's images sometimes felt posed or overly edited. Skin texture and depth of field often looked more natural with Midjourney. ChatGPT images often had a slightly warm, yellow tone that wasn't present in Midjourney's outputs.

Even with more specific portrait requests, Midjourney often had better aesthetics. But this is where differences started to appear.

Hands and Human Details

One area where AI models often struggle is hands. In portrait close-ups, ChatGPT's hands often looked good and anatomically correct. Midjourney has improved, but issues still appeared, especially when hands were not the main focus.

We tested complex hand poses:

• Asking for a specific chord on a guitar: Neither model got this right. Fingers were often misplaced.

• Holding up specific numbers of fingers on different hands: ChatGPT nailed this. Midjourney struggled, sometimes only showing one or two fingers total.

• Rock, Paper, Scissors poses: ChatGPT understood the game and created the correct hand shapes. Midjourney did not.

Overall, ChatGPT proved more reliable at generating correct hands when asked for specific configurations.

Prompt Understanding and Following Instructions

This was a big test. While Midjourney often produced beautiful images, it frequently missed specific details requested in the prompt. ChatGPT was much better at including every element, no matter how small.

Examples:

• Complex portrait details (lighting, clothing, specific tattoos, scars): ChatGPT got nearly every detail perfect. Midjourney missed several.

• Objects in specific arrangements (dog sitting on a cube, spoon on an apple): ChatGPT followed these instructions precisely. Midjourney sometimes got parts right but missed others.

• Highly complex scenes (chess board with specific tile types and piece descriptions): ChatGPT came much closer to generating the described pieces. Midjourney struggled significantly with the piece details.

If your prompt needs to be followed exactly, ChatGPT has a big advantage right now.

Generating Text

Generating readable text in images is a known difficulty for many AI tools. Midjourney V7 showed very little improvement from previous versions. For anything beyond very simple words, the text was often jumbled or nonsensical. ChatGPT, however, was incredible at generating correct and readable text based on prompts, even for complex phrases, slogans, and names. For any task requiring text integration, ChatGPT is the clear winner.

For many businesses, getting images that follow brand guidelines or include specific text is crucial. Automating these tasks can save massive amounts of time and effort. Did you know you can automate image generation tasks including prompt adjustments and variations? Check out the Midjourney Automation Suite from TitanXT to streamline your workflow and get precise results faster.

Faces in Crowds and Distant Shots

Midjourney still struggles with faces in images, especially when there are many or they are far away. Pictures of concerts or busy streets from Midjourney often had morphed or unnatural-looking faces when zoomed in. ChatGPT handled crowds and distant faces much better, producing more consistent results.

Censorship and Replicating Likeness

We tested limits on generating copyrighted characters or images of public figures. Midjourney was generally more relaxed with big-name intellectual property like Disney characters. Both had limitations on what public figures could be shown doing, though ChatGPT was slightly less censored overall.

When trying to generate less famous people or replicate a likeness from a photo for people not generally known, Midjourney struggled a lot. The explore page of Sora (which uses the same DALL-E model as ChatGPT) showed it can create very good likenesses, even from low-quality photo prompts, for a wide range of celebrities.

Different Styles

Both models can produce impressive images in many styles, like cinematic, candid, tilt-shift, mixed media, and even anime. ChatGPT can do Studio Ghibli style. It struggled with some other famous anime artist styles but could replicate them if asked by movie title instead.

Midjourney often had better aesthetics for certain styles, especially abstract or surreal art. However, again, this was conditional on whether Midjourney correctly followed the prompt, which was less likely for specific or complex requests.

Consistent Characters

A major benefit for ChatGPT right now is its ability to create multiple images featuring the same character. Midjourney V7 currently lacks the character reference tools that were available in Version 6. An 'omni reference' tool is planned, but until then, ChatGPT is the clear winner for projects needing character consistency across multiple scenes.

Imagine needing an image series with the same character but different poses, outfits, or settings. Doing this manually with Midjourney is difficult. With a tool designed for automation, generating variations while keeping consistency becomes much easier. Explore what's possible with the Midjourney Automation Suite by TitanXT. Handle variations, aspect ratios, detailed prompts, and more automatically.

Default Creativity

We also tested giving the models very short, even one-word prompts to see what they would create without much direction. This tests their default creativity. Overall, Midjourney seems to have a lead here, often generating more surprising or aesthetically striking images from minimal input. It excels when you want abstract, weird, or purely aesthetic results.

Which AI Tool is Best?

It depends on what you need:

• Choose Midjourney V7 if: You prioritize raw aesthetics, especially for abstract or surreal art, and don't need extremely specific details or text in the image.

• Choose ChatGPT if: You need excellent prompt adherence, readable text in images, consistent characters across multiple shots, good hands/anatomy, or reliable faces in crowds.

Midjourney is generally faster for quick iterations, especially with its Draft mode where you describe a vibe and it creates the prompt and image quickly. ChatGPT is much slower at image generation.

Both tools have editors and other features, though Midjourney's current editor is more robust. However, comparing the core ability to generate images from prompts, each has clear strengths and weaknesses.

Conclusion

Neither Midjourney V7 nor ChatGPT is perfect, but they excel in different areas. Midjourney wins on overall aesthetics and default creativity. ChatGPT wins big on prompt adherence, text generation, and character consistency. The best tool depends entirely on the specific task you're trying to achieve.

Ready to take your AI image generation workflow to the next level? Stop wasting time on manual prompting and variations. The Midjourney Automation Suite from TitanXT can handle complex tasks and batches, freeing you up to focus on the creative vision.