AI Photo Creation
Myrt's GRIP homework for Monday evening using ChatGPT40-mini (Pro account.)
To this point, I have shied away from creating images using AI (Artificial Intelligence). But part of taking the GRIP Genealogy Institute AI course is doing homework using information gleaned each day. Please note I’m using primitive prompts but am doing better at conversing back and forth with the LLM to arrive at a desired image.
ChatGPT 04-mini
PROMPT: Create an image of a man sitting at his desk studying something on his laptop. Include a stack of old books and random printouts from his printer. Use the ANIME style with shades of blue and chartreuse. A window to the side would provide added interest.
RESPONSE: I’m sorry, but I wasn’t able to generate the image. Let me know if you’d like to try again or adjust the request.
PROMPT: Can you explain why you were unable to generate the image?
RESPONSE: The image-generation service returned an error when I called the image_gen.text2im tool—it didn’t complete the render. That can happen if the tool experiences a transient internal failure or if its servers are temporarily unavailable. Because it didn’t produce an output, I can’t show you the illustration. Let me know if you’d like to try again or adjust your requirements.
PROMPT: Please try again.
RESPONSE: (after generating the 1st iteration) Here’s the anime-style illustration of a man at his desk studying on his laptop, complete with a stack of old books, scattered printouts, and a window to the side. Let me know if you’d like any tweaks!
PROMPT: (to modify the 1st iteration image generated) Please remove the paper on the wall in the background and replace it with a framed family tree.
RESPONSE: (after generating the 2nd iteration.) I’ve updated the illustration by removing the paper on the wall and replacing it with a framed family tree. Let me know if you’d like any further adjustments!
PROMPT: How about changing the style of the illustration to a circa 1890s-style image, replacing the laptop with a typewriter?
RESPONSE: (after generating the 3rd iteration.) I’ve transformed the scene into an 1890s–style ink and watercolor illustration, with the laptop replaced by a manual typewriter. Let me know if you’d like any further adjustment.
PROMPT: Can you make this look like a photographic image?
RESPONSE: (after generating the image below.) Here’s the converted realistic photographic image. Let me know if you’d like any further tweaks!
SUMMARY
Pleased to note that ChatGPT40-mini is not hallucinating by adding a 6th finger on each hand. And I’m comfortable adding information to tweak the look of the images.
Happy family tree climbing!
Myrt :)
DearMYRTLE,
Your (not so) retired friend in genealogy.
Sometimes I think ChatGPT plays games with me. When I was generating an image of two men playing cards in a tavern recently for one of my posts it returned a Pic that was perfect in every detail except the cards in one man's hand were back to front. He was looking at the back of them. I asked it to turn them round but leave everything else the same. It did but one man had cards with a blue back like the stack of cards on the table and the other (the one it turned around) had a red back. I asked it to fix that and leave everything else the same but in addition to doing that it added a ridiculous hat to what of the men. Feeling slightly exasperated I gave up for the day. I finally got something acceptable the next day.
Interesting! I like the last one, which really looks like a photograph. I've tried a few times but didn't like my results. I think I'll go play a bit with it now.