A New Era for Character Consistency in AI Art: Midjourney’s Game-Changing –cref Property
Apr 07, 2024![An AI-generated image of a woman with red hair sitting on a park bench looking at the camera.](https://kajabi-storefronts-production.kajabi-cdn.com/kajabi-storefronts-production/file-uploads/blogs/2147749292/images/a7a3e2e-23b-18bf-4668-fbdcad8168e2_cref-header.png)
Hello Tech Savvy Friends,
If you’ve ever played with Midjourney or Dall-e, you have probably come across the frustration of generating consistent characters.
Innovators have tried to find a way around this, from naming the character to having AI generate multiple blocks of the character in one image. None of these were great, nor were they considered actual workarounds.
Maintaining character consistency across multiple scenes is one of the main hurdles needed to achieve less of a “slot machine” return and more intentional results. Midjourney introduced a new property a few weeks ago, and I am pretty impressed with the results.
The introduction of the “character reference” property marks a pivotal moment in AI artistry. It offers creators opportunities to weave more cohesive and intricate narratives with AI images.
The Challenge of Character Consistency
One downside to using AI to generate art has been the struggle to keep a character’s appearance consistent across different scenes and contexts. Characters would morph unexpectedly from one image to the next, disrupting the visual appeal and making it impossible to tell long-form stories with a single character.
Enter –cref: Midjourney’s Solution
Midjourney’s introduction of the –cref property represents a leap forward in tackling this issue. This feature allows creators to reference a specific character and render that character throughout various scenes. This ensures that the character retains its core visual identity, regardless of the setting or action. Essentially, –cref acts as an anchor, guiding the AI to remember and apply key characteristics of the referenced subject across different generative scenarios.
Results: Grows and Glows
This technology excels in several areas, and like any innovation, there are still some growth areas to be explored.
Let’s take a look at four headshots I generated a month ago. I wanted to use these four distinct women and see if I could generate them in various locations with various emotions and surroundings. The outcome generated here is a combination of two to three returns each. You are seeing the best of those multiple returns.
Prompt: A woman sitting on a park bench reading.
Prompt: A woman standing in the rain wearing a yellow jacket and holding a red balloon on a city street.
Prompt: A woman, upset and sad, sitting at the coffee shop.
Prompt: A woman playing in the living room with her kids, the room is bright and airy.
As you can see above, I have noticed that rendering multiple people in one scene makes it very difficult to get a photorealistic version of the characters.
Below, I tried using a combination of two of these women in one scene. I was not able to achieve the results I consider successful.
As impressed as I am with the rendering of people, I have not had the same luck with animated characters. Now, I have not had a lot of time to play, so maybe I will pick up some tricks along the way.
Implications for Storytellers and Artists
This development has profound implications for artists, illustrators, and storytellers who leverage AI in their creative processes. With the character reference property, narratives can unfold with a newfound depth and continuity, enabling visually cohesive stories and characters that grow more familiar and relatable over time.
For Graphic Novelists, Comic Creators, Picture Book Authors
Graphic novelists, comic creators, and picture book authors, in particular, stand to benefit immensely. Maintaining character consistency opens the door to more ambitious storytelling, where visual continuity plays a crucial role in narrative depth and engagement.
In Animation and Concept Art
Animators and concept artists can use this feature to experiment with characters in various environments, testing different backstories, emotional states, or action sequences without losing the essence of the character’s design. This can significantly speed up the concept development phase, allowing for rapid iteration and exploration of creative ideas.
Personalized Content Creation and Marketing Materials
For those creating personalized content or marketing materials, the property ensures that key branding elements or characters can be seamlessly integrated across an entire campaign, providing a consistent visual thread that ties all pieces together.
AI Fear For Artists
The concept of Jevon’s Paradox, recently discussed by Seth Godin, presents a thought-provoking counter to the fears surrounding technological advancement, particularly regarding AI and artists. The paradox suggests that as technology improves efficiency, instead of leading to decreased use, it paradoxically increases demand for that very resource. This counterintuitive outcome has profound implications for understanding the evolving dynamics between human work, technological advancement, and AI.
Image by Sketchplantations
Godin’s insight that “when AI learns to write computer code, the demand for programmers goes up” illuminates an optimistic perspective on AI’s role in the future of work. As AI technologies make certain tasks more efficient, they don’t necessarily render human skills obsolete; rather, they elevate the demand for those skills by making the outcomes they produce more in demand.
This view challenges the dystopian narrative that AI and automation will lead to a loss of jobs. Instead, it suggests that AI could catalyze the creation of new opportunities, jobs, and industries that we cannot yet fully envision. Just as the human race has created billions of jobs within a single lifetime due to technological and societal advancements, AI’s contribution to efficiency could start a similar explosion of new roles and professions.
How To Use
Simply put the –cref property at the end of the Midjourney prompt, then insert a URL for the reference image. It should look like this: –cref URL. You can also refer to more than one image by putting two URLs back to back or separating them with a comma. I have not had much luck with this one yet.
Conclusion
Midjourney’s property is a paradigm shift in how we approach AI-generated art. It empowers creators to tell richer, more coherent stories, building worlds that captivate and characters that endure.
This is indeed a new digital art and storytelling era, one where consistency meets creativity, opening up a universe of possibilities for creators everywhere.
Let’s embrace this journey, for in Midjourney’s new world, our characters can finally travel with us, unchanged, from one adventure to the next.