#13 We are finally able to create decent consistent characters within AI-generated content
An exciting week lies behind us with a lot of big updates. In this week’s Substack we’re talking about Devin (an AI software engineer), generative UI instead of AI and the update of Midjourney who finally announced consistent character functionality. Let’s go!
Devin, the first AI software engineer
Cognition, an applied AI lab focused on reasoning, has unveiled Devin, the inaugural AI software engineer with the capability to develop code, websites, and software from a simple instruction, aimed to complement the efforts of human engineers.
What is Devin?
Devin is an AI-powered software engineer capable of planning and executing complex engineering tasks that typically require thousands of decisions. Equipped with advanced long-term reasoning and planning capabilities, Devin can work independently or collaborate seamlessly with human engineers. By recalling relevant context at every step, learning over time, and fixing mistakes, Devin sets a new standard for AI-assisted software development.
What Can Devin Do?
Devin's capabilities are vast and diverse. From learning unfamiliar technologies and building end-to-end applications to autonomously finding and fixing bugs in codebases, Devin can handle a wide range of software engineering tasks. Some notable examples include:
Creating an interactive website simulating the Game of Life.
Devin can train AI models by itself.
Devin can autonomously find and fix bugs in codebases.
Gaining Early Access to Devin AI
Interested individuals can apply for early access to Devin AI by completing a submission form on the official Cognition Lab website: https://www.cognition-labs.com/
Generative UI (and it’s not a misspell): Component generator Vercel
Vercel, a leading platform for frontend developers, has recently introduced v0, an innovative product that simplifies website creation by combining the power of generative AI with best practices in frontend development. This new technology, called Generative UI, allows developers to create interfaces by simply describing their ideas. Sounds wild right?
v0 has already been used by various developers to build the first iterations of their products. For instance, here it is utilised to create the initial version of a Resume Builder product, while Braintrust, a new AI platform, built its pricing page using v0. These examples showcase the versatility and effectiveness of v0 in helping developers quickly bring their ideas to life.
Prompt: ”Sign up page” or “A profile card for a user in the middle screen”
Easy to use
Using v0 is a straightforward process for frontend developers. To get started, simply describe the interface you want to build, and v0 will generate code using popular open-source tools like React, Tailwind CSS, and Shadcn UI. You can then select an iteration and continue editing within the v0 environment. Once you're satisfied with the result, copy and paste the generated code into your app and continue developing from there.
v0 offers both free and paid subscription plans, allowing developers to choose the option that best suits their needs and budget.
Midjourney consistent characters
Midjourney, the popular AI image generation service, has recently launched an exciting new feature that allows users to recreate characters consistently across multiple images. This breakthrough addresses one of the major challenges faced by AI image generators to date.
How it works:
1. Users can apply the "--cref" (character reference) tag at the end of their text prompts in the Midjourney Discord server.
2. Following the tag, users paste a URL of a previously generated Midjourney image featuring the character they want to recreate.
3. Midjourney will then attempt to match the character's facial features, body type, and clothing from the referenced image in the newly generated image.
4. Users can control the degree of similarity between the new image and the original character by using the "--cw" tag followed by a number between 1 and 100. A higher number results in a closer match to the original reference.
prompt --cref {refURL} --cw 0 ⎜ face only
prompt --cref {refURL} --cw 25 ⎜ face & hair
prompt --cref {refURL} --cw 50 ⎜face, hair, some clothing
prompt --cref {refURL} --cw 100 ⎜face, hair, and clothing
5. Multiple characters can be combined into a single image by using two "--cref" tags side by side with their respective URLs.
This new feature is a significant step forward for Midjourney, as it allows users to maintain narrative continuity in their creations, such as storyboarding for films, novels, graphic novels, or comic books.
Some users have been testing it with characters other then people, such as with cars. Which seem to work as well: