👋 Hey and welcome to AI News Daily. 

Each week, we post AI Tools, Tutorials, News, and practical knowledge aimed at improving your life with AI. 

_{Read time: 4 minutes}

Table of Contents

• Intro

• Why consistent characters matter

• Mastering Midjourney's Character Reference: A Guid …

  • How it Works

• AI ANNOUNCEMENT OF THE WEEK

• 🎯 HOT TAKES

Intro: Bridging the Consistency Gap in AI-Generated Imagery

Midjourney, a leading service in AI-driven image creation, has rolled out a highly requested update: the capability to consistently replicate characters across multiple generated images.

Addressing this issue marks a significant step forward in the field of AI imagery, where ensuring character uniformity has been a notable challenge. This challenge originates from the reliance of AI image generators on “diffusion models”. 

These models, including those similar to Stability AI’s Stable Diffusion, operate by interpreting textual prompts from users and constructing images bit by bit. The process is informed by a vast database of human-made images and associated text descriptions, a method that, until now, has made the consistent depiction of characters a complex task.

Thanks for reading AI News Daily! Subscribe for free to receive new posts and support my work.

Why Consistent Characters Matter — The Challenge with AI-Generated Art

The magic of generative AI, seen in tools like OpenAI’s ChatGPT, lies in their ability to spawn unique creations from each prompt. This feature, while exciting, introduces a notable challenge: the inconsistency of outputs. Despite entering the same or similar prompts, the AI conjures up entirely new content every time.

This trait is invaluable for generating fresh pieces of work, such as the diverse images Midjourney produces. However, it poses a dilemma for creators aiming to maintain character consistency across various scenes in storytelling mediums like films, graphic novels, or comics. Achieving narrative continuity, where characters retain their appearance and personality traits across different settings and expressions, has been a steep hill to climb for generative AI.

Midjourney steps into this gap with an innovative solution: the “–cref” tag, short for “character reference.” This tool allows creators to anchor their characters' identity across multiple creations, attempting to preserve specific attributes like facial features, body type, and attire through a URL linked in their prompts.

As Midjourney refines this feature, it's setting the stage for the platform's evolution from an experimental tool to a valuable asset for professional creatives, offering a solid foundation for consistent character portrayal in AI-generated imagery.

How to use the new Midjourney consistent character feature

This innovative feature excels when used with characters previously crafted within Midjourney. For instance, users should start by creating or locating the URL for a character they've already generated.

Let’s start from scratch and say we are generating a new character with this prompt: “curly-haired blonde woman in a pink dress”.

Select the preferred image to upscale, then right-click it within the Midjourney Discord to copy its link.

Next, input a fresh prompt like “wearing a hat and standing in a forest –cref [URL],” inserting the URL of your original character image. Midjourney will then endeavor to recreate your character within this new context.

While the outcomes might not precisely align with the initial character or prompt, the similarities observed are promising.

Furthermore, users have the opportunity to influence how closely the new rendition mirrors the original character by using the “–cw” tag along with a value ranging from 1 to 100 at the end of their new prompt (like “–cref [URL] –cw 100”). A lower “cw” value means more variation from the original, whereas a higher value aims for closer replication.

–cw 100:

–cw 8:

Combining multiple characters into a single image is also possible by employing two “–cref” tags, each followed by its respective URL.

Launched this week, this feature is already being explored by artists and creators. If you're a Midjourney user, give it a go. Read founder David Holz’s full note about it below:

Mastering Midjourney's Character Reference: A Guide to Consistency and Creativity

How it Works

• Simply add "--cref URL" right after your prompt, where the URL points to an image of your character.

• To adjust how closely the new image should resemble the reference, use "--cw" with numbers from 100 to 0.

• By default, setting "--cw 100" ensures the AI focuses on the character’s face, hair, and outfit.

• At "--cw 0", the emphasis is solely on the face, perfect for when you want to switch up outfits or hairstyles.

Purpose of the Feature

• Best results are seen with characters generated by Midjourney. Real-life photos might not translate as well, similar to standard image prompts.

• "Cref" zeroes in on character traits much like regular prompts but with a sharper focus.

• Don't expect it to catch super fine details like specific facial marks or logo designs on a t-shirt.

• Compatible with both Niji and the usual Midjourney models, and it can also work in tandem with "--sref".

Advanced Features

• Blend traits from different images by using more than one URL, like "--cref URL1 URL2", akin to combining multiple images or styles in prompts.

Usage on the Web Alpha

• Dragging or pasting an image into the imagine bar now shows three icons to categorize it as an image prompt, a style reference, or a character reference. Holding Shift while selecting allows an image to fulfill multiple roles.

Conclusion

Since MJ V6 is currently in the alpha stage, there may be some adjustments to its features. However, we're eagerly awaiting the release of the beta version. Once it's fully developed, the need to train SD for consistent character results will be eliminated.

The --cref feature is primarily intended for use with actual photographs, where its effectiveness is somewhat limited. While it can capture basic features, achieving an exact match is rare. However, when a more stylistic, abstract approach is taken rather than strict photorealism, this feature shines as a powerful narrative aid.

PS: I curate this AI newsletter every week for FREE, your support is what keeps me going. If you find value in your reading, share it with your friends by clicking the share button below!

AI ANNOUNCEMENT OF THE WEEK

The Devin AI announcement has taken the spotlight as this week's most talked-about tweet. It has sparked a range of reactions: content creators are voicing their frustrations through YouTube videos, developers are being challenged to showcase their talents, and chatbot enthusiasts are riding a new wave of excitement for coding applications.

Key Highlights:

• Devin excels in complex engineering tasks, making thousands of decisions, learning from experience, and autonomously correcting errors.

• It transcends its initial programming, capable of adapting to new technologies.

• Devin integrates smoothly into existing projects, matching a human engineer's proficiency and actively engaging with user feedback to refine its approach.

• In the SWE-bench coding benchmark, Devin resolved 13.86% of real-world issues, a significant leap from the previous 1.96% benchmark.

• Demonstrating exceptional versatility, Devin has successfully passed engineering interviews at top AI firms, fine-tuned models with minimal guidance, developed and deployed applications, and even completed projects on Upwork by navigating various tools autonomously.

🎯 HOT TAKES

• Building an AI Salesman in Under 10 Minutes 

• Five Tips To Set the Right Mindset for New AI Startups 

• Easiest Way To Make Money On YouTube Using AI