MidJourney AI Art: Comparing Version 1 to Version 4 Image Generation
Table of Contents
- Introduction
- Sword Image Comparison
- Cyberpunk Cityscape Comparison
- Knight Image Comparison
- Dragon Image Comparison
- Wizard in Dungeon Comparison
- Beautiful Woman Portrait Comparison
- Beholder from Dungeons and Dragons Comparison
- Sword in Art Nouveau Style Comparison
- Dragon in Art Nouveau Style Comparison
- Dragon Carved in Marble Comparison
- Necromancer Portrait Comparison
- Neon Female Paladin Comparison
- Conclusion and Summary
Introduction to MidJourney AI Art Generator
MidJourney is an AI image generator that creates images based on text prompts. It uses a technique called diffusion, which starts with random noise and gradually transforms it into a coherent image matching the prompt.
MidJourney is currently on version 4, which was released in November 2022. Each version brings improvements in image quality and capability. In this post, we will compare images across versions 1 through 4 to highlight the rapid progress.
Overview of Key MidJourney Features
Some of the key features of MidJourney include: • High-quality images tailored to prompts - The AI learns from vast datasets to generate surprisingly good images that match text descriptions • Support for styling through modifiers - You can append words like 'watercolor style' or 'cinematic lighting' to transform the aesthetic • Quick iteration with Upscale - Each image starts low-res, then quickly sharpens on command to save computation • Community pooling of credits - Members donate credits into a common pool enabling free generation for those without a subscription MidJourney strikes an excellent balance between quality, flexibility, and accessibility compared to alternatives.
Image Quality Improvements in Version 4
Version 4 brings significant leaps in image quality over previous versions: • More accurate forms and compositions • Greater level of detail with less abstraction • Improved coherence especially for complex prompts • Enhanced handling of stylistic elements like lighting and texture As we will see through image comparisons, Version 4 stretches the imagination with its fidelity and artistry compared to the past.
Sword Image Comparison Across Versions
Prompt: 'A sword'
Version 1 produces an abstract shape vaguely recognizable as a sword. Version 2 aligns the blade and hilt better but still appears more abstract. Version 3 oddly encases the sword in another form.
Version 4 is a massive jump up in realism. Beautiful lighting and detail emerge despite the simple prompt. Form, lighting, texture and style occur naturally without any guidance.
Cyberpunk Cityscape Comparison
Prompt: 'A cyberpunk city'
Version 1 has a pleasing composition but is quite abstract. Version 2 improves contrast and angle. Version 3 tries a new composition but loses intensity.
Yet again, Version 4 is photorealistic right out of the gate. Neon lights, towering skyscrapers, and hazy smog capture the cyberpunk aesthetic. Tiny flying vehicles dot a futuristic skyline.
Knight Image Comparison
Prompt: 'A knight'
Version 1 only vaguely gets at armor pieces. Version 2 loses all semblance of a helmet. Version 3 recovers the helmet but stays moderately abstract.
Version 4 generates striking armor full of excellent details down to the emblem on the breastplate. The only oddity is the unusual hand pose.
Dragon Image Comparison
Prompt: 'A dragon'
Version 1 gets at wings and a neck but little else. Version 2 similarly fails to capture head or limbs. Version 3 resembles a snake more than a dragon.
Version 4 clearly depicts a roaring dragon in flight even if the neck bending seems improbable. The quality achieved invalidates all previous attempts.
Wizard in Dungeon Comparison
Prompt: 'A wizard in a dungeon with dimensional smoke'
Version 1 shows a blurry indistinct scene. Version 2 better highlights the wizard against the background. Version 3 conveys smoke but no clear wizard.
Version 4 perfectly realizes the scene down to the wizard's stoic expression. Fine smoke swirls around a cave-like environment. Details emerge well beyond the prompt specifications.
Beautiful Woman Portrait Comparison
Prompt: 'A beautiful woman'
Version 1 distorted the mouth and eyes. Version 2 corrupted both face symmetry and features. Version 3 exhibits a disconcerting left-right mismatch.
In Version 4, MidJourney finally resolves age-old facial generation difficulties. Symmetry aligns perfectly and features appear natural. The quality advances enable creative portraiture.
Beholder from Dungeons and Dragons Comparison
Prompt: 'A purple beholder from dungeons and dragons'
Version 1 includes inconsistent eyes and loses the color. Version 2 stays orange and fails to depict the iconic central eye.
Version 3 recovers purple color and one eye but the form stays difficult to parse. Version 4 conveys an instantly recognizable beholder with details like fangs and skin texturing.
Sword in Art Nouveau Style Comparison
Prompt: 'A sword in art nouveau style'
Version 1 aptly conveys the graceful curves but the sword form gets lost. Version 2 tries more detail but departs from the intended style.
Version 3 takes an entirely new direction. Version 4 ignores style guidance but the ornate hilt stands on its own as a visual delight.
Dragon in Art Nouveau Style Comparison
Prompt: 'A dragon in art nouveau style'
Version 1 lacks discernible form. Version 2 hints at wings in the cluttered composition but fails to come together.
Version 3 could depict any number of organic shapes. Only Version 4 clearly conveys the prompt with a dragon curled amid the signature Art Nouveau curves and botanical shapes.
Dragon Carved in Marble Comparison
Prompt: 'A dragon carved out of marble'
Neither Version 1 nor 2 come anywhere close to reasonable approximations. Version 3 gets out a baby dragon head shape at best.
Version 4 skillfully generates an ornate dragon head sculpture befitting ancient ruins. Veining even mimics natural marble patterns.
Necromancer Portrait Comparison
Prompt: 'A necromancer portrait painted by greg rutkowski and alphonse mucha in a floral art nouveau watercolor style'
Version 1 rather hauntingly distorts eyes and nose in an otherwise interesting aesthetic. Version 2 stays stylistically appealing but face symmetry falls apart.
Version 3 makes progress but facial discrepancies still unsettle. Version 4 finally enables a properly proportioned figure complete with staff, gloves, and background scene.
Neon Female Paladin Comparison
Prompt: 'A female neon paladin portrait painted by artgerm and greg rutkowski in a CGI style'
Versions 1 and 2 both fail in representing high-tech armor or coherent faces. Version 3 conveys neon tones but facial distortion dominates.
Version 4 delivers a perfect execution of the prompt down to color scheme, attire, pose and backdrop. Tiny finishing touches take the image beyond expectations.
Summary and Conclusion
Across 14 diverse prompts with varying complexity, MidJourney Version 4 overwhelmingly surpasses all previous iterations in generating striking, nuanced images.
Prompts once doomed to vague shapes or distorted features can now produce realistic representations even matching specific styles. Lighting, texture and detail get conveyed as further enhancements.
Still, Version 4 has limitations including odd anatomical irregularities and failure to perfectly execute some stylistic guidance. But the magnitude of improvements enables creations usable for professional digital art purposes.
We eagerly await future MidJourney updates to see just how far AI image generation can push the boundaries of imagination.
FAQ
Q: What is MidJourney?
A: MidJourney is an AI image generator that creates images from text prompts. It leverages deep learning and generative adversarial networks.
Q: What versions of MidJourney were compared?
A: Versions 1 through 4 of MidJourney were compared using the same text prompts.
Q: What kinds of images were generated?
A: Images like swords, landscapes, portraits, and fantasy images were generated for comparison.
Q: What was the result of the image comparisons?
A: Version 4 showed clear improvements in image quality, detail, and accuracy compared to earlier versions.
Q: Did version 4 perform well on complex images?
A: Yes, version 4 created impressively detailed dragons, wizards, beholders from Dungeons & Dragons.