Diana the Adventurous Reporter 11÷20

Pages 11 ÷ 20 of my graphic novel. In the previous blog of this series I told you about the difficulties you may encounter when you want to build a storyboard, a graphic novel or a comic book with GPT Image 2, especially if you don't want to let the model control the development of the story, with all the possible errors that this procedure entails, but you need to maintain complete professional control over each panel: not only the characters, their clothes, the background, but also the captions, speech bubbles, and text. If you look at this series, you will notice that from page 11 (and from pages 1-10 that I published in the previous blog post) to page 20 there is a progressive and tangible improvement. When creating the first series with GPT using the images I already had available as 9 reference images (image-to-image), I had to write a very long prompt (the model allows up to 5000 characters), specifying what could be kept and what had to be changed in each panel, and what had to be written in each caption and speech bubble. Each page took me over an hour, and once I got the result, I had to accept that the layout of panels, captions, and speech bubbles was entirely determined by GPT; sometimes there were errors in the text transcription (not present in my prompt) or in the placement of the tails of some speech bubbles. Attempting to correct these errors and improve the quality of the page took time, credits, and a lot of patience, and often new images where a bug had been fixed contained something incorrect that wasn't present in the previous version. I finally decided to solve all these problems once and for all. GPT is perfect if you want to create amateur comics using a few characters of your choice as references, describing the development of your story in the prompt and letting the program decide what to show in each panel (accepting possible errors). But if you want a precise development of the storyboard it is advisable to create the images for each vignette in advance. In this regard, GPT is very useful, because it allows you to easily create consistency, for example, between a character's clothing in different panels. Using GPT, I was able to change page 20 from this version, where Diana's and the old man's outfits are different in different panels: to this one, where they are very consistent: To achieve reliable results, however, you need to use pages with panels without captions or speech bubbles. In a future blog, I'll explain in detail the most correct, fastest, and most effective procedure to adopt if you'd like to try your hand at this type of work. To be continued...