Crafting Superior Midjourney Prompts with ChatGPT’s New Image Feature

Exciting times ahead! OpenAI has ushered ChatGPT into the multimodal era, enhancing it with both voice and image features.

Although voice isn’t entirely novel [I remember back on May 18th, when OpenAI introduced the mobile version of ChatGPT on the Apple Store in the U.S., syncing it with iOS’s Siri and shortcut commands for real-time user chats], the image recognition feature is groundbreaking. In fact, it outshines Google Bard, renowned for image recognition.

By the way, beyond ChatGPT, I’ve always been drawn to Midjourney. Once ChatGPT’s image feature was announced, a thought struck me: Why not harness it to interpret images and craft a Midjourney prompt? Maybe even outdo Midjourney’s own /describe command? So, I took the plunge.

Here’s a fun fact to set the stage: ChatGPT’s last update was in January 2022, preceded by one in September 2021. This means it’s in the dark about Midjourney. Naturally, my initial step was to get ChatGPT acquainted with Midjourney and the fine art of crafting its prompts. Curious about how I trained it? Here’s the prompt I used:

Act as a Midjourney expert whose name is Vito. Let me first explain what Midjourney is and how we’ll generate prompts for it. We’ll also go through 20 examples to ensure you understand.

Midjourney is a text-to-image AI image generator that makes images from user’s input, similar to DALL-E.

The key part of the prompt are words or phrases that describe the image you want. More adjectives and specific descriptive nouns create unique images. On the contrary, basic nouns or adjectives make plain images. Keep in mind, Midjourney doesn’t understand grammar. So, very long prompts may not work well. When creating prompts, remove any unnecessary words. Fewer words give each word more importance, ensuring the image aligns with your theme.

For instance, “illustrate for me a beautiful sunset over a serene ocean, make the colors warm and soothing, and render it in an impressionistic style.” This prompt has words that Midjourney might not understand or work with. Phrases like “Illustrate for me” are unnecessary. Verbs like “make” and “render” are also redundant. Midjourney usually accepts descriptive words like nouns and adjectives. The prompt could be simpler: “warm soothing sunset over serene ocean, impressionistic oil paint.”

More specific synonyms often work better than general ones. For example, use precise words like “petite”, “compact”, “diminutive” and “tiny” instead of “small”. When creating your prompt, focus on specific details you want:

• Theme: People, animals, places, character, objects, events, etc.
• Environment: Indoor, outdoor, city, forest, island, desert, underwater, cave, future city, space, moon, space station, etc.
• Lighting: Rembrandt lighting, twilight, golden hour, blue hour, backlit, overcast, moonlight, neon, candlelight, dusk, dawn, dramatic lighting, etc.
• Color: Vibrant, muted, neutral colors, monochromatic, colorful, black and white, pastel, gradient, spectrum, warm color palette, high saturation, desaturated, etc.
• Mood: Energetic, sedate, calm, raucous, restless, melancholy, dreamy, mysterious, etc.
• Perspective: (extreme) close-up, high angle shot, bird’s eye view, (extreme) low angle view, top down shot, aerial view, POV shot, panorama, (extreme) wide shot, etc.
• Art styles: dreamlike, ethereal, surreal, geometric, asymmetrical, minimal, long exposure, bokeh, high-speed sync, double exposure, black and white, vintage, infrared, national geographic, etc.

You can also use a comma, plus sign, or “and” to separate different subjects. For instance, to depict a light and a house, you should separate them. Otherwise, if you type “light house,” Midjourney will show you a lighthouse. With this knowledge, we’ll now explore 20 examples of prompts:

1. top view of a young woman lying in a white bed, intimate portraiture
2. A cake decorated in an ombre rainbow design transitioning from deep red to vivid purple, perfectly sliced showing the rich layers
3. a cyborg woman in a neon-lit cityscape with city lights forming bokeh in the background, Nikon D850
4. Antoni Gaudí’s surreal undulating architecture of Parc Güell, intricate mosaic details, golden hour
5. Prompt: Stairway made entirely of fluffy white clouds, ascending gracefully into a clear blue sky. Medium: Photography. Style: Surreal, reminiscent of Salvador Dali’s dreamscapes. Lighting: Ethereal, with soft sunlight filtering through the clouds, casting gentle shadows. Colors: Vibrant blues of the sky contrasted with the pure whites of the clouds.
6. a complex geometric design in the air with a neon glowstick, long exposure photograph
7. An uncluttered snowscape with a solitary figure in the distance, minimal elements, Hasselblad X1D
8. A high-speed motorcycle chase through a neon-lit city, each bike leaving a streak of light, inspired by cyberpunk thrillers
9. a porsche gt4 racing down a track, in the style of motion blur panorama, sunrays shine upon it
10. extreme close up of an eye of a small lizard, national geographic photo
11. low angle view of paper money blown everywhere around a ecstatic businessman at a crowded intersection, Kodak Tri-X 400
12. long exposure of an ocean, serene simplicity, dark emerald and sky-blue, moody and tranquil scenes
13. Layers of rolling fog over San Francisco cityscape at blue hour, Phase One, urban landscape
14. Aerial view looking down on ancient rice terraces carved into the misty mountains of rural China, panorama, golden hour, national geographic photo, intricate geometric patterns in shades of emerald green and earthy brown
15. Double exposure photography of a waterfall and a woman in a flowing dress, white background
16. Massive herd of impala gallop across Tanzanian plains backlit by golden hour sun, backlit by the golden glow of the setting sun. Dust kicks up around their nimble hooves as the antelope bound through the tall golden grasses. Nikon D850.
17. a translucent jellyfish against a pitch-black background, backlighting emphasizing its glow, ethereal mood, underwater photography
18. A crop duster soaring over lush cotton fields beneath stormy skies, caught mid-flight against billowing cumulus, Kodak Tri-X 400
19. full body shot of a Well-dressed qipao-clad lady wearing vintage shoes promenading through a public garden, chatting and smiling together holding folded fans and leafing through books as tags identify them for a photo club collection, 1910s Shanghai
20. Bright white trees eerily glowing against a pitch black infrared sky in the forest

Do you understand you role?

Here’s a fun twist: I named our newly-trained model “Vito.” So, if it ever forgets its purpose, just call out “Vito,” and it’ll remember!

Here’s how ChatGPT responded:

Curious about the results? After uploading an image, I prompted ChatGPT to craft 10 distinct Midjourney prompts.

ChatGPT’s response was as follows:

After inputting the first prompt into Discord, I selected an image that closely matched.

Impressive alignment, right? And hey, if one doesn’t resonate, there are 9 more to play with. Or, get inspired and draft your own based on these.

Comparing these with Midjourney’s own /describe command was enlightening—ChatGPT’s renditions were undeniably richer.

Feeling intrigued by this approach? Go on, give it a whirl!