Imagen 3 vs Midjourney: A Visual Face-Off

midjourney blog post image

First Impressions: Imagen 3 Within Gemini

Imagen 3 is now part of Gemini. You simply ask Gemini to create a picture. It’s good that Google’s image modeling is finally integrated, but I’ve long appreciated Midjourney’s distinctive artistic style.

There are some limitations with Imagen 3. The free version can’t create images of people, but the advanced version can. Also, images are always square. No landscape options here. Let’s see how it compares to Midjourney’s more stylistic output.

The Core Comparison: Style and Authenticity

I’ll be using Midjourney prompts, which may give it a slight advantage, since the prompting style and artistic direction differ. Midjourney has a unique artistic style. Some images feel genuinely created. Other AI image generators lack that artistic flair, making them seem synthetic. While great for photographs, they often lack authenticity. But is this really true of the new Google model?

Portrait Test: Coffee Bar Selfie

The first test was a photographic portrait using the prompt: “coffee bar selfie of a Japanese woman shot in a low-quality phone.” Here are the results:

  • Midjourney: Produced a very realistic, selfie-like image. It captured the low-quality phone aesthetic well. It doesn’t immediately scream “AI generated.”
  • Imagen 3: Also created a great image, complete with a stylish haircut. The arm position looks natural, as if she’s holding a phone. Perhaps I’ve been too harsh on Imagen 3. It seems the examples I’ve seen before didn’t use detailed prompts. The key might be in the prompting!

The square format of the image works well for a selfie style. But what about more artistic styles?

Artistic Styles: Graffiti Influence

Next, I used a graffiti-style prompt, referencing Chris Foss. The goal was to see how well each generator could capture artistic chaos and detail. The prompt was “image of cool guy in style by Chris F graffito influence Urban core rage core cartoonish humor”.

  • Midjourney: The result had a distinct artistic style, complete with a fake signature. The image included graffiti-like lines and textures, giving the impression of a graffitied wall.
  • Imagen 3: Produced clean lines, but lacked the artistic chaos of Midjourney. It didn’t quite capture the cartoony element. Midjourney excels at capturing art styles here.

Midjourney seems to have the upper hand when it comes to artistic styles.

Album Cover Recreation: The 1980s Test

I asked both to create an album cover from the 1980s. I included lyrics from a song I created in the Midjourney prompt.

  • Imagen 3: The layout felt more like a 1960s Chinese portrait. It included some text, possibly Japanese, with the words “Osaka girl.” It also recreated the old imprint of the album from underneath. The haircut feels very 1980s American.
  • Midjourney: Recreated the medium the picture is on, adding crease lines, giving it the feel of a worn LP cover you might find in an antique record shop.

While the Imagen 3 result was good, it wasn’t as stylish as the Midjourney version. There’s something very believable about the Midjourney output. For lecturing or consulting on AI art, this is the kind of image that would impress people.

Pushing the Boundaries: Illustration for a Movie Poster

I used a detailed description to generate a movie poster illustration. The description included specific artists and techniques. The prompt was “illustration for a movie poster illustration by renie grow and JC liend Decker and Conrad ronet using wordcloud bibl graffiti typography organically integrated Bolder angles experimental framing unusual perspectives”.

  • Imagen 3: The result was stylish and interesting, though I’m not an expert in the referenced art styles. The image felt more AI-generated and had a 1950s feel.
  • Midjourney: The result had a 1960s feel. Excellent image overall.

Skateboard Design: Bauhaus Inspiration

My son wanted to create a Bauhaus-inspired skateboard. I used this as inspiration for both platforms. The prompt was “picture of a bow house inspired design skateboard”.

  • Imagen 3: Gave a straight image, seemingly placing it in an art gallery setting. The skateboard included realistic details like scratches and dirt.
  • Midjourney: Meant to create a design breakdown, and it delivered a very good image. But the Gemini’s was exceptional.

Where Gemini excels is in portraiture. If you put in some detail, you will get great results.

Godzilla: The Pencil Drawing Test

The prompt was “create a color pencil drawing of Godzilla in the style of detailed penciling position drawing heavy shading of Godzilla”.

  • Imagen 3: The result was decent but felt a bit “AI-ish.”
  • Midjourney: Looked more like actual pencil work, with great detail. The output was much more artistic.

Revisiting the Past: A Portrait from 2 Years Ago

I recreated one of my first portraits from Midjourney (about 2 years ago) to see how the platforms compare now. I used the exact same prompt: “photograph of Port beautiful Japanese woman of 30 smiling blah blah blah”.

  • Imagen 3: The hair may need some work, but it looks like a portrait. It feels like a corporate photo. It’s hard to tell if it’s truly black and white.
  • Midjourney: (From 2 years ago) Has a lot of detail, shading, and lighting. The skin texture and spots are visible. The result feels like it was done by a professional photographer trying to capture character. Also more artistic.
  • Midjourney: Produced four different styles. Some had weird textures. However, some were very painterly. The models are much more interesting than the Google result.[/LI
  • Imagen 3 (take 2): The result was generic, but realistic. If you want a straight, boring, vanilla man drinking coffee, it delivers.

Artistic Recreation: Woman on a Beach

I tried recreating a “beautiful Japanese woman on a beach splatter fashion” image from February 2024.

  • Midjourney: The result was artistic and very cool. The different styles were impressive.
  • Imagen 3: Seemed to interpret “splatter fashion” literally, as if something was splattered ON the fashion. It lacked an understanding of the artistic styles.

The Final Portrait: Indian Village Woman

Finally, I prompted a “portrait of an Indian village woman in forest in himal parades Pradesh clear facial features cinematic 35 mm lens.”

  • Midjourney: Captured the lighting and reflections well. The wrinkles and hair were detailed, and the lens use created a great bocker effect. Very good.
  • Imagen 3: Went with a younger woman. It lacked the accent lighting and global illumination. Still a great portrait, but feels a bit darker.

Midjourney’s lighting was better and captured more of the intended artistic style. The Gemini image feels more photojournalistic, while the Midjourney one feels more like a coffee table book photograph.

Final Verdict

Imagen 3 isn’t bad for creating images. They’ve come a long way from the bad AI images of the past. For vanilla, generic, but great-looking images and photographs, especially, Imagen 3 is fine. But if you’re looking for that artistic edge, Midjourney still has it. I am impressed with the realism of some of the photography-style images that Imagen 3 is creating.

What do you think? Have you used Imagen 3? Let me know in the comments!

🖥

Do This Locally with Logen

Skip the subscription. Generate images like these on your own GPU — private, unlimited, no cloud dependency.

Try Logen Free →

Midjourney vs Logen

FeatureMidjourneyLogen
Cost$10-60/moFree & Pro plans
PrivacyCloud (images stored)100% local
GenerationsLimited by planUnlimited
Internet RequiredAlwaysNo
Custom ModelsNoYes (any ComfyUI model)
GPU RequiredNo (cloud)Yes (local)
Ease of UseDiscord botDesktop app
Try Logen Free →
FREE 11-PAGE PDF HANDBOOK + LOCAL AI FIELD NOTES

Get the Free 11-Page Local AI Image Generation Handbook

See what GPU VRAM you need for FLUX.1-schnell and SD 1.5, how checkpoints and LoRAs work, and five repeatable workflows on Windows and Linux x64. Enter your email to open the 11-page PDF right away and get our six-part weekly local AI field notes.

6–24 GB VRAM Tables Cloud vs. Local Math 5 Studio Workflows
How signup uses your email

We register your email with the TitanXT newsletter on beehiiv to send the 11-page PDF link, our six-part weekly local AI sequence, and occasional Logen updates. You can unsubscribe in one click from any email or by writing to support@titanxt.io. Read beehiiv's privacy policy (new tab).

Scroll to Top