Diana Harper the Adventurous Reporter 1 to 10

For all of you who don't visit my page, here are the first 10 tables of my graphic novel featuring Diana, the adventurous reporter. To be continued... I hope you enjoy the story, and will continue to follow it. Now, since I apparently have to write at least 360 words, otherwise the blog can't be published, I'll tell you about the pros and cons of GPT Image 2 for creating graphic novels and comics. GPT is quite effective when using text to image, describing the development of the story panel by panel, or even as a general outline, leaving the model free to interpret it and arrange the panels as it sees fit. You can also use images of the main characters in the story as reference images, asking the model to maintain consistency in facial and body features and clothing throughout the various panel, and describing the poses, interactions, and environment of each scene in the prompt. However, I created Diana's story over two years ago as a picture story, meaning each picture had its own caption describing how the story unfolded. At the time it was not so easy to maintain character consistency across dozens or even hundreds of images. In particular, the characters' outfits changed very easily. But in the end, with the help of a few tricks, I managed to obtain a series of about 500 images that, while not perfect, allowed everyone to follow the story. Now I thought it would be easy for me to turn Diana's story into a graphic novel using GPT in image to image mode, and each picture in the story as reference image for the corresponding panel. I was hoping GPT could easily improve and enhance the images, making them more consistent when needed. But this turned out to be only partly true. Not infrequently GPT confused the content of one panel with that of another, even though each panel and each reference image were numbered unambiguously. Furthermore, sometimes the image quality was not improved at all, and the reference image was of higher quality than the corresponding vignette made by GPT. As for captions, speech balloons, and the cutting and positioning of each individual panel, GPT often did what it wanted without taking into account the suggestions in the prompt. So, as I gradually progressed through the creation of the tables, I had to develop a technique to speed up their creation and make the results more reliable and less random. But maybe we'll talk about that another time!