Today I ran some fast experiments to take a closer look at the Euler a Sampling Method, the various Sampling Step options and how each impacts a very simple AI Text Prompt’s output. I used Stable Diffusion Version 1.5 web interface provided by AIGeneration.blog for everything mentioned in this article. You can add your email to the wait list to get invite access here.

Step One: Brainstorm AI Text Prompt Ideas

I used the DDIM Sampling Method because it runs very fast and allows me to quickly verify whether my AI Text Prompt idea will work as I hope. Speed allows me to iterate on each of my ideas very quickly until I design an AI Text Prompt that works as I need it to. I tested a variety of brainstorming ideas from complex to simple and finally ended up happy with a very simple text prompt: “1970s Crime Drama ” using the DDIM Sampling Method at 10 Steps, again for 50 Steps, 100 Steps and 150 Steps with nice results that were not too dramatically different from previous step number choices. This is a great option for fast iterative brain storming.

Step Two: Euler_a, Crime Drama

Next, I took my final chosen AI Text Prompt from my DDIM Brainstorming experiments and used it with the Euler_a Sampling Method. Euler_a usually outputs high quality for the subject matter I am working with today (Character portraits of human actors starring in specific film genre).

The different number of Euler_a Sampling Steps produce dramatically different output results from each but they are each great quality within the 10 to 75 step range. Euler_a does run slower than DDIM, so I tend to use DDIM for fast brainstorming and then use Euler_a for slower high quality AI Art Generations with more photo real characters, environments and actions.

The following images show the process and AI Art Generations:

AI Art Generation	AI Text Prompt:	Euler_a # of Sampling Steps:
	web interface for AI Art Generation app we use here at AIGeneration.blog. Join the waitlist to access it when available for more artists.
	1970s Crime Drama I’m happy with the layout, colors and style of this. See how a small number of steps with Euler_a can produce decent output?	10
	1970s Crime Drama This result is dramatically different from the previous result while remaining relevant to what I asked for and of decent quality.	150
	1980s Crime Drama The faces are garbled but I like the colors, lighting, shadows and format of this. It’s useable with typical post processing tasks.	10
	1980s Crime Drama I like this output, its good enough for web projects as is and of course can be improved with post processing tasks. Our AI Generation app provides Face Restore, UpScale and more to make post processing easier and quicker.	100
	1990s Crime Drama This is an interesting assemblage. The eyes and faces can be improved with our Face Restore options.	10
	1990s Crime Drama I’m happy with this just as it is. I like the expressions on their faces.	50
	2000s Crime Drama Again, I like the faces, expressions, style and colors.	50
	2000s Crime Drama I like this but do not like that garbled text shows up on the generations as I increase the number of Sampling Steps with Euler_a in this experiment.	75
	2000s Crime Drama The original output and then I clicked the ‘Face Restore’ button to improve the eyes.	150
	2020s Crime Drama Interesting result that reminds me of using our AI Art Generation services to design film storyboards, comic books and similar digital / print assets.	10
	2020s Crime Drama Another story board esque result. it’s interesting that this result mixes photo real with red / white / black hand painted cartoon style in the same output.	50

So far my experiments have provided some interesting results. It was very noticeable that setting the step numbers to larger numbers actually reduced the quality of the AI Art Generations with garbled faces and alot of garbled text. I did not include the garbled gobbly gook in the table above because I prefer to show the output that is decent and hope that readers can get access to the stable diffusion web interface on AIGeneration.blog to play and experiment with their own ai text prompt ideas.

Step 3: Futuristic Crime Drama –no-text

I added a new parameter to my ai text prompt because I noticed that ‘Futuristic Crime Drama’ was producing output with garbled text in every Sampling Step number with Euler_a. Adding the ” –no-text” to the very end of my simple text prompt is producing the results I like and will show next.

AI Art Generation	AI Text Prompt	Euler_a : # of Steps:
	Futuristic Crime Drama, –no text	10
	Futuristic Crime Drama, –no text	10
	Futuristic Crime Drama, –no text	20
	Futuristic Crime Drama –no text I removed the comma to look at any differences in output	50

I’m going to stop today’s experiments here and plan our next articles to focus on the next set of questions and requests on how to do specific tasks with the AIGeneration.blog AI Art Generation web service.

Let me know if you want to share your own experiments and art that you make with our AI Art Generation web service and I’ll put you in my schedule to be featured on this website.

Subscribe to be the first notified when our AI Art Generation app is available for more people.

Tag me on your social media posts (@monigarr) about AI Art Generation and please do share links to our articles on social media so I can talk about all things AI Generation with you too.