As we begin 2024, the progress in AI-powered image generation is truly impressive. The capabilities of AI to generate, enhance, and modify images have evolved significantly, presenting a range of possibilities for various sectors, including literature and publishing.
In recent times, we’ve seen AI progress from a tool that could simply apply filters or imitate artistic styles to a valuable resource capable of creating images from textual descriptions, extending images, and even modifying specific elements within an image. The boundary between human-made and AI-generated art is becoming increasingly indistinct, and we’re enthusiastic about being a part of this evolution.
Our mission is driven by a passion for literature and a desire to harness these advancements to refresh old public domain books. These books, often forgotten or overlooked, are rich with stories and ideas. However, their charm can sometimes be diminished by outdated or missing illustrations. This is where AI can play a role.
In 2024 and beyond, we plan to employ the most recent image generation AI tools, such as OpenAI’s DALL-E 3 and Outpainting algorithms, to create preliminary versions of updated artwork for these books. These AI tools, with their capacity to generate images based on textual descriptions and extend images, offer an excellent starting point.
However, our process doesn’t end with AI. We believe in the power of human creativity and judgment, and that’s why we utilize Adobe Illustrator and Photoshop to refine and perfect the AI-generated images. This ensures that the illustrations not only align with the style and tone of the book but also meet our quality expectations.
As we embark on a year that promises further technological advancements, we’re excited about the potential of AI to aid us in modernizing more old public domain books. By merging the power of AI with human creativity, we hope to rejuvenate these books, making them more accessible and appealing to today’s readers.
As we delve deeper into the digital age, the application of Artificial Intelligence (AI) has permeated various fields, including the realm of illustration and image generation. AI has the potential to breathe new life into old, public domain books by transforming their illustrations into vibrant and engaging visuals. In this blog post, we will explore some of the most prominent image generation AIs and their potential applications.
DeepArt and DeepDream, developed by Google, are two of the most well-known image generation AIs. DeepArt transforms your photos into works of art using the styles of famous paintings, while DeepDream uses a convolutional neural network to find and enhance patterns in images, creating dream-like, psychedelic images.
These tools could be used to revamp the illustrations in old books, giving them a modern, artistic twist. Imagine an old children’s book being reimagined with the style of Van Gogh’s “Starry Night” or a classic novel adorned with dream-like, surrealistic illustrations.
DALL-E, developed by OpenAI, is an AI program that generates images from textual descriptions. It’s capable of creating images of objects that don’t exist in the real world, making it a powerful tool for illustrators and graphic designers.
DALL-E could be utilized to create new illustrations for old books based on their textual content. This could be particularly useful for books where the original illustrations have been lost or damaged, or for books that were never illustrated in the first place.
Runway ML brings the power of machine learning to creatives, allowing them to use AI models in their work without requiring any coding knowledge. It offers a wide range of models for different tasks, including image synthesis, style transfer, and object detection.
For instance, illustrators could use Runway ML to apply a specific artistic style to all illustrations in a book for a consistent aesthetic, or to generate new images based on specific themes or elements in the book.
GANPaint Studio, developed by MIT-IBM Watson AI Lab, is an AI-powered image editing tool. It uses Generative Adversarial Networks (GANs) to allow users to add, modify, or remove specific elements in an image.
With GANPaint Studio, illustrators could modify the original illustrations in old books, adding or removing elements to better align with the modern aesthetic or to update outdated or offensive elements.
OpenAI’s DALL-E 3, an enhanced version of the previous DALL-E models, has been a game-changer in our process of rejuvenating old public domain books. This AI model generates images from textual descriptions with unprecedented precision and creativity. We’ve been using DALL-E 3 to create initial versions of the updated artwork, using descriptions directly from the books to generate images that adhere to the original author’s vision.
The key to DALL-E 3’s enhanced prompt-following abilities lies in training the model on highly descriptive, generated image captions. Earlier text-to-image models often struggled with detailed image descriptions, frequently ignoring words or misinterpreting the meaning of prompts. This issue largely stemmed from inaccurate and noisy image captions in the training dataset.
To address this, we trained a custom image captioner and used it to recaption the training dataset. By training several text-to-image models on these synthetic captions, we found a reliable improvement in their prompt-following ability. This led to the development of DALL-E 3, a new text-to-image generation system that has shown to perform favorably in terms of prompt-following, coherence, and aesthetics when compared to its competitors.
Alongside DALL-E 3, we’ve also been utilizing Outpainting algorithms. Outpainting, or image extrapolation, extends the content of an image in a semantically consistent manner. This technique has proven to be a powerful tool for creating full-page illustrations based on the initial images generated by DALL-E 3.
Despite the remarkable capabilities of AI tools, human creativity and expertise remain vital in the process of creating compelling and cohesive illustrations. This is where Adobe Illustrator and Photoshop come into play.
Adobe Illustrator, a renowned vector graphics editor, allows us to fine-tune the AI-generated images, adding details, adjusting shapes, and ensuring the illustrations match the style and tone of the book. It provides a level of control that AI currently cannot match, enabling us to fine-tune the illustrations to our exact specifications.
Adobe Photoshop, a leading raster graphics editing tool, is used to adjust colors, apply filters, and add effects to the AI-generated images. This helps us ensure that the illustrations not only match the text content but also evoke the appropriate emotions in the readers.
In summary, the integration of AI-generated initial designs and human-led refinement ensures that the updated artwork remains true to the spirit of the original books while appealing to modern aesthetics. It’s a blend of old and new, of human and machine, resulting in illustrations that are both nostalgic and innovative. The use of AI in illustration is not about replacing human artists but rather about providing them with powerful tools to enhance their creativity. By using AI and traditional design software in tandem, we’re able to breathe new life into old public domain books in a way that respects the original work and excites today’s readers.
