How to Create Multiple Consistent Characters in Midjourney V6!

The video explores the possibility of generating consistent characters in Midjourney. The narrator has tested Midjourney's new consistent character feature, aiming to create images featuring two characters: a woman in a yellow shirt and a man in a blue turtleneck. The results exceeded expectations, so the video will explain the process and share tips and tricks. First, the narrator generates reference photos for the characters: a Black woman in a yellow collared shirt and an Asian man in a blue turtleneck. The prompts for these images specify using Kodak Portra 400 film for consistent style. To maintain character consistency, the narrator uses the /prefer_option_set command to save the reference photos under custom names ('Lisa' and 'Kim') in Midjourney. These references are then easily invoked in prompts using --Lisa and --Kim, ensuring that Midjourney uses the correct reference images for each character.

The narrator moves on to generating base images for the characters, explaining a specific prompt format that includes the camera angle, setting, and character descriptions, followed by the camera film stock. The first example features a low-angle shot of a woman in a yellow button-down shirt and a man in a light blue turtleneck, both captured using Kodak Portra 400 film. The narrator ensures that the character descriptions match the reference photos, keeping the photo styles consistent. Midjourney generates a base image matching the prompt, which the narrator uses to inject reference photos of the characters. The narrator demonstrates using the Vary Region inpainting tool to replace the face of the man with the reference photo of Kim and adjusts the prompt to use --cref and --kim for consistency. The resulting image shows a well-matched facial structure, although the hairstyle needs improvement, which the narrator promises to address later.

The video continues with another example, highlighting the importance of considering skin tone consistency when injecting character images. When injecting Lisa into an image, the narrator notices a slight difference in skin tone between the reference photo and the base image, particularly on the woman's hands. By selecting both the woman's head and hands during inpainting, the narrator achieves a more consistent skin tone. Attention to detail in aspects like skin tone is crucial for consistent characters. However, hairstyles remain a challenge. The narrator shows an example where Kim's hairstyle is satisfactory, but Lisa's is not. The issue arises because Lisa's original hairstyle has dreads in a bun, but the inpainted image lacks the bun. By adjusting the selection area to include more of the desired hairstyle, Midjourney can produce an image closer to the reference. This approach helps achieve the desired hairstyle, but some scenarios require additional prompts to achieve consistency.

Addressing the persistent hairstyle challenges, the narrator demonstrates another example involving a high-angle shot of two people walking down the stairs. Despite attempts, the hairstyle generated for Lisa doesn't match the reference due to the extensive selection area needed. To overcome this, the narrator modifies the prompt to explicitly specify desired hairstyles: dreads in a bun for Lisa and a fade for Kim. The generated image now features hairstyles much closer to the original references. The narrator advises selecting the region around Lisa's head for inpainting and tagging her with --cref to maintain character consistency. This method significantly improves hairstyle accuracy, providing a solution to one of the common challenges in generating consistent characters.

The video shifts focus to Niji style anime images, questioning if consistent character generation is achievable for this style. The narrator provides reference photos for a man with orange hair and eyes and a woman with purple hair and eyes, both in the style of 'My Hero Academia.' The importance of maintaining a consistent anime style is emphasized due to variations in appearance across different anime backgrounds and characters. The narrator attempts to create a base image with these characters in an ancient Buddhist temple. Despite following the same method, the result is unsatisfactory. Although some similarities exist, the faces don't match the references well enough. The narrator concludes that achieving multiple consistent characters in anime style art requires further refinement and experimentation. This guide, being advanced, invites viewers to explore additional resources for a comprehensive beginner introduction to Midjourney's consistent character feature.