For this assignment, I was tasked with choosing a prompt to sketch, describe what it should look like, and generate a few variations from the prompt. I choose the prompt “Fly-fishing for rainbow trout in Montana” because I like fishing, and was curious how a detailed landscaped would look like if it was generated by an AI.
The prompt I used was “Man wading in a river next to a forest fly-fishing for rainbow trout on a sunny day in Montana”, which I think is descriptive enough for an AI service to generate an image with.
I used deepai, an online image generator that’s available without creating an account, or purchasing a subscription. I submitted the prompt with the default options and created this image in 3 seconds:
I was satisfied with the result, and wanted to find out how different changes affect the scene. I want to make the person wear a red coat, and have the weather change to snow, so I changed the prompt to “Man wearing a red jacket wading in a river next to a forest fly-fishing for rainbow trout on a snowy day in Montana”.
This seems stylized like a classic painting, even though the settings were kept the same. The water on the bottom has weird reflections, which is noticeable, but can be edited out really quickly.
Now I want to see if I can add a fish to the image, and use a specific type of tree in the background. The new prompt I used is “Man wearing a red jacket wading in a river next to an autumn forest fly-fishing, successfully catches a large rainbow trout on a snowy day in Montana”.
None of the trees are autumn-colored, and there’s no fish to be seen. It’s not necessarily bad, but it needs a decent amount of editing to change what the ai was supposed to produce.
Deepai was overall, serviceable for simplistic prompts, but adding more details to the prompt confuses the AI into interpreting certain details in unintended ways. I wouldn’t personally use this particular program to create complicated scenes, which have way too many errors to edit in a reasonable amount of time which could be used to generate different images. It would definitely be worth it to purchase a subscription to a more famous service for actual use, but for quick mockups for the price of nothing, Deepai was just good enough.
2 replies on “Blog Assignment #1b – AI Image Generation”
Hey Keith. I was unable to see the AI generated pictures in your post, whether opened in Chrome or Safari. Can you make sure they are in a format that is visible to us? Otherwise, clearly you did the assignment, and explored the tool well.
Sorry, I didn’t see this comment until today. The images and my opinions on them were recently corrected.