AI Images.
The topic that I am choosing to explore with an AI image generating tool for this blog assignment fly-fishing for rainbow trout in Montana. I recently took a trip to Montana, and my grandpa used to be a fly fisherman. This is a nostalgic and romanticized image in my mind.
I have used AI image generator tools in the past and know that they work very well if you have a particular style in mind. One of my favorite styles to experiment with AI in is pixel art. I love pixel art because it reminds me of old video games that I love and especially loved when I was younger. The AI image generator that I am going to use for this assignment will be Dalle3 through microsoft bing image creator.
I want this image to look like a pixel art rendition of a stereotypical picturesque scene of fly fishing on a river in the forested mountains of Montana. Someone would be fly fishing in the foreground wearing waiters and be equipped with a pole, net, hat, and pack. They will be standing in a river that meanders around a bend into the horizon. The banks of the river will have evergreen trees. And tall mountains shaped by the river valley will be in the background. It will be a partially sunny day with fluffy white clouds in the sky. The fly fisherman will be excited to have just landed a rainbow trout on the line.
This is my sketch of my idea:
My prompt for my first set of generated images was this, “Pixel art of a boy fly fishing for rainbow trout in Montana. The boy is wearing waders and a hat and is equipped with a fly fishing rod, net, and pack. The boy is standing in the river. The river follows a bend into the horizon. There are evergreen trees on the banks of the river. There are mountains in the background. It is a partially sunny day with fluffy clouds in the sky. The boy is excited because there is a rainbow trout on the line.” Here are the results:
Results from the first round were pretty on point with what I was hoping to receive. The image generator basically nailed the setting and the outfit of the main character. The primary issues involved the fish not looking like it was being caught, and the boy is standing on the bank of the river, rather than in the water. I adjusted the prompt to try and address these issues.
This is my prompt for the second round of images, “Pixel art of a boy fly fishing for rainbow trout in Montana. The boy is wearing waders and a hat and is equipped with a fly fishing rod, net, and pack. The boy is standing knee deep in the water. The river follows a bend into the horizon. There are evergreen trees on the banks of the river. There are mountains in the background. It is a partially sunny day with fluffy clouds in the sky. The boy is casting the fly fishing rod while rainbow trout swim by.” Here are the results:
The second round produced much of the same results as the first image, with the same issues. I altered the prompt more dramatically for the third round.
This is my prompt for the final round of images, “Pixel art of a boy reeling in a rainbow trout on his fly fishing rod in Montana. He is standing in the water. The river is up to his knees. The boy is wearing waders and a hat and is equipped with a fly fishing rod, net, and pack. The river follows a bend into the horizon. There are evergreen trees on the banks of the river. There are mountains in the background. It is a partially sunny day with fluffy clouds in the sky.” Here are the results:
This time around the boy was finally standing in the water, and the image is almost exactly what I am looking for, however the fish still doesn’t quite look right. It does not look like it is being reeled in, in the way it would in real life.
Overall I am very happy with the results. There are still quirks that would likely take a little while of prompt refinement, but it is so much faster than trying to create that myself, and so much cheaper and faster than commissioning an artist to create it for me.
The Image Generator through Microsoft Bing with Dalle3 is a very impressive tool. The design is really clean and straightforward, which is why I prefer it over the competition which often looks messy or over the top. Every prompt is responded to with one to four options, and is seemingly quite successful at creating what you have in mind. Some limitations include that it is hard to iterate. It would be nice if you could generate an image using text as images as the input, as opposed to just text- or at least select one of the resulting images and refine. Also there are only 15 free images available to generate a day. In some ways this is a pro and a con. More access is desirable, but a truly free option is very appreciated even if it’s limited it still includes all the important features.