- 1,250 views, 2 today
165
I recently got access to a neat tool called MidJourney AI. It's a bot that allows you to put in specific prompts and get images generated off that prompt. Here are some examples of what I would consider "high quality" work based on the prompts:
These pieces generated content accurately to the prompt, VERY WELL. Though, through this process, I've found that there is a "sweet spot" where if you include too many details about the work it's a hindrance. Below are a few images that didn't generate properly.
Through these, it's become clear that being too vague produces crappy results (scary Waluigu was supposed to be "Pirate Waluigi") but going to complex produces equally crappy results (see the frog things. Those were supposed to be DnD Characters). Also, certain specifications and details don't work with certain things. For example, with the frog things, I gave it the prompts of Cinema 4D, Redshift, Octane Render, Photoreal, and other things of that nature. Most of these qualifiers are going to pull data from images of simulations of liquid, smoke, etc. So not really the greatest for photoreal.
Through trial and error, I've found that the following things have a MAJOR impact on performance.
1. Base Idea
This is going to be the first thing the AI sees and the core of your generated content.
2. The Style
The style should be inspired by an artist or a time period (can be multiple things)
3. The Aspect Ratio
With the base aspect ratio being a square, you're not locked to that. Think about it like a box - the AI is going to generate something that fits in that box, however, if you made the box wider it's going to change the content to be better suited for that space.
Here are some Good/Interesting generations along with the prompts
Base Idea = Green
Style = Blue
Aspect Ratio = Yellow
So, based on these I've come to the following conclusion.
AI is not going to replace artists any time soon. however, I think this could be a great tool for helping artists brainstorm and visualize crude ideas with little development.
If you want me to try out some prompts, comment below!
Good AI
These pieces generated content accurately to the prompt, VERY WELL. Though, through this process, I've found that there is a "sweet spot" where if you include too many details about the work it's a hindrance. Below are a few images that didn't generate properly.
Bad Ai
Through these, it's become clear that being too vague produces crappy results (scary Waluigu was supposed to be "Pirate Waluigi") but going to complex produces equally crappy results (see the frog things. Those were supposed to be DnD Characters). Also, certain specifications and details don't work with certain things. For example, with the frog things, I gave it the prompts of Cinema 4D, Redshift, Octane Render, Photoreal, and other things of that nature. Most of these qualifiers are going to pull data from images of simulations of liquid, smoke, etc. So not really the greatest for photoreal.
Through trial and error, I've found that the following things have a MAJOR impact on performance.
1. Base Idea
This is going to be the first thing the AI sees and the core of your generated content.
2. The Style
The style should be inspired by an artist or a time period (can be multiple things)
3. The Aspect Ratio
With the base aspect ratio being a square, you're not locked to that. Think about it like a box - the AI is going to generate something that fits in that box, however, if you made the box wider it's going to change the content to be better suited for that space.
Here are some Good/Interesting generations along with the prompts
Base Idea = Green
Style = Blue
Aspect Ratio = Yellow
Good AI with Prompts
Space Freighter, in the style of The Outer Worlds Video game Peter Elson Jim Burns, Epic 1970s, --ar 16:9
demonic teddy bear, in the style of Tim Burton Anato Finnstark, epic volumetric lighting, --ar 9:16
Pigeon, in the style of Jim Steranko Sean Phillips Marvel Comics, --w 1920 --h 1080
demonic teddy bear, in the style of Tim Burton Anato Finnstark, epic volumetric lighting, --ar 9:16
Pigeon, in the style of Jim Steranko Sean Phillips Marvel Comics, --w 1920 --h 1080
Extra Prompts (And Usernames)
So, based on these I've come to the following conclusion.
AI is not going to replace artists any time soon. however, I think this could be a great tool for helping artists brainstorm and visualize crude ideas with little development.
If you want me to try out some prompts, comment below!
Tags |
1 Update Logs
Update #1 : by MisterGriimm 06/08/2022 1:33:23 pmJun 8th, 2022
Generated images based off the usernames of the people who have diamonded the post so far.
fishstacks
-Wolfix-
Adrestio
dreamCritting
fishstacks
-Wolfix-
Adrestio
dreamCritting
tools/tracking
5602116
6
ai-generated-art
Create an account or sign in to comment.
Gotta say I LOVE @Dreamcrittings prompts, very magical.
I really like the 4th one that was generated for your username. The structure looks like it was built out of a deceased god with a winding pathway. So cool