Midjourney impacted hugely the stock photo and design industry. But it still needs to rely heavily on references and celebrity images to generate a consistent character or model. However, I have tried to overcome this challenge and develop some ideas. While not perfect, it’s a close approximation.
Steps
First, you can use prompts that include various angles of characters (front and side views) and set the aspect ratio to a maximum of 7:4 or 2:1 (–ar 7:4)
https://docs.midjourney.com/docs/aspect-ratios
Then, use the “Vary (strong)” and “Vary (Subtle)” options to get more variations of the same person from the desired output. Generating more variations takes time and patience, as well as comparing them.
Once you have chosen a similar-looking image, upscale it and click “Vary (Subtle)” Next, choose two similar-looking pictures and combine them using a screenshot or editing software. This will be your reference image.
Upload your reference image to Discord and copy the image URL address. Finally, add the reference image to the prompt below, including the image URL, prompt, front and side view shots, an aspect ratio of 2:1, and style raw with a version of 5.2.
Image prompts:https://docs.midjourney.com/docs/image-prompts
Tips: Refine your results using the image weight parameter and “Vary (Subtle)” option. This process allows you to create unique, consistent images that meet your needs.
You can use Photoshop “generate fill” AI to enhance the images.
You can create similar-looking characters by working on the variations and generating more images. I want to share some of the results I have obtained.