People Are Using Camera Filenames to Make Midjourney More Photorealistic

Matt Growcoot
Three-panel image: Left shows a graduate in cap and gown with a man, middle displays a scenic view of a rocky forest path along a river and mountains, right depicts a smiling couple posing closely together in casual attire.
All of these AI-generated images were made on Midjourney by using a camera filename in the prompt.

Midjourney released its latest V7 model last week and already users have been inputting creative prompts that include a filename from a camera system to achieve photorealistic AI images.

One X user, who goes by Min Choi, published a thread of images in which he writes a generic camera filename before a descriptive prompt. Choi chose CR2, Canon’s previous generation RAW image file format. For example, Choi used the prompt “IMG_7249.CR2” followed by “casual selfie at X.”

The results are astonishing, producing images that are almost impossible to detect as synthetic.

A woman with long wavy hair wearing a blue headband, sunglasses on her head, and a casual summer outfit smiles and makes a peace sign. She's standing on a wooden walkway by a sunny beach, with blue skies and clear water in the background.
Min Choi
Two smiling young women pose together indoors at a nighttime event. One wears a black dress with a necklace, and the other wears a dark, sparkly top. There is a dimly lit background with an exit sign above a partially visible person.
Min Choi
A smiling couple wearing sunglasses takes a selfie in a hallway. The woman has blonde hair in a bun, wearing a floral dress. The man has a beard and a tank top with a design. Both seem happy and relaxed.
Min Choi
Two people smiling happily at a night event. The woman on the left has curly hair and is wearing a black dress. The man on the right is wearing a white shirt. They're surrounded by others, indicating a lively atmosphere.
Min Choi
A woman with long, curly hair smiles while taking a selfie in a cozy room. The room is softly lit with string lights, and a patterned tapestry is visible in the background.
Min Choi
A person with short, wavy dark hair and freckles is taking a selfie in an office. They're wearing a light-colored button-up shirt. In the background, an office setting with furniture and another person is visible.
Min Choi

It’s not just Canon filenames either, Levi Forster used the iPhone image format HEIC, in his Midjourney prompt.

“AI is trained on photos,” Forster explains. “Many real photos have file names like IMG_02202021.HEIC — so if you feed it just a typical file name in the prompt and that’s it, it will generate a realistic looking picture because the data that corresponds is always real.”

A Reddit user has been trying out this method for the last few months in a bid to create realistic looking photos on the AI image generator.

“CR2 file is a RAW image file format created by Canon digital cameras,” the Reddit user explains. “The name CR2 stands for Canon Raw 2. The number part I’ve just made up so you can mess around with the numbers but just make sure it’s in that order.”

Users can simply enter a generic camera filename in the prompt, nothing else, and Midjourney will come back with “something random.”

A pint glass filled with amber-colored beer topped with a frothy white head sits on a wooden table in a cozy pub, with softly lit background featuring bar stools and warm lighting.
Salmontunabear
A hand holds a waffle cone topped with a scoop of vanilla ice cream swirled with pink and yellow hues. The background shows a softly blurred window, suggesting a bright day.
Salmontunabear
A slice of cheese pizza sits on a weathered wooden picnic table outside. The background is blurred, showing leafless trees and a cloudy sky, indicating an autumn or winter setting. The table is slightly wet from rain.
Salmontunabear
A scenic mountain trail lined with tall, sparse trees and large rocks, leading into lush, wooded hills. A narrow stream flows alongside the path under a bright, clear sky, with sunlight filtering through the branches.
Salmontunabear

Meanwhile, other users have praised the new model’s “base level realism”, showing off photorealistic examples.

Midjourney’s V7 model is currently in its alpha phase and is still rolling out all of its features. It has a new “Draft Mode” which is reportedly 10 times the speed and costs less. The catch is that it generates images in lower-resolution.

,
, , , , ,
PetaPixel articles may include affiliate links; if you buy something through such a link, PetaPixel may earn a commission.
Related Articles
Midjourney Bans AI Images of Chinese President Xi Jinping
dog You Can Now Buy and Sell AI Image Prompts on a Marketplace
Midjourney v5 example Midjourney v5 Creates Photorealistic Images and Even Does Hands Correctly
vicotrians Victorian-Era People Who Never Existed: These Portraits Were AI-Generated
Discussion