Start creating free

Changing character & style with Image to Image

@CaylaCatz@CaylaCatz··4 min read·30 comments
Changing character & style with Image to Image

Let's start at the end so you know where we are going.

What I wanted was this: a goth artist female, the journal, the quotes. But I didn't want to type all those quotes. I will show you how I go them.

Illustration: Changing character & style with Image to Image

I started with the idea when @DavidP introduced me to a new word : solipsistic in his post

https://budgetpixel.com/p/86930

The word looked cool so I checked it out.

Solipsistic

Solipsistic is an adjective from solipsism . According to Wiki:

Solipsism (/ˈsɒlɪpsɪzəm/ SOLL-ip-siz-əm; from Latin sōlus 'alone' and ipse 'self') is the philosophical position that only one's own mind is certain to exist. It holds that knowledge of anything outside one's own mind, including the external world and other minds, is uncertain and cannot be conclusively known.

https://en.wikipedia.org/wiki/Solipsism

I thought was a cool word so I made a very simple prompt:

🖍️✒️🖍️. solipsistic

That's my prompt. I just wanted to see what an AI model would do with the word. The emojis and the word. The emojis 🖍️✒️🖍️ in some models results in an artistic style like color pencils, crayons, or an inked graphic novel style. I got this from :

Muse 1.0

Illustration: 🖍️✒️🖍️. solipsistic

I love what Muse did. I like the quotes on the wall, the journal and more writings. Muse created the quotes based on the tenets of solipsism philosophy. They are very cool.

The 🖍️✒️🖍️ did not result in an art style but instead the image has cryons and a pen in it. Plus it's probably what resulted in the art journal.

So i thought I'd like the same thing but with a female goth artist and in one of my favorite ink styles.

So I made a 2nd prompt. I started the prompt with:

Muse 1.0 model

🖍️✒️🖍️. solipsistic goth female artist dressed in lacepunk purple

+ my styleblock: no kanji. no hanko. silver rinpa school. use a felt-tip brush pen to ....

and then I added one of my favorite styles which says a bit about ink going from thin delicate lines to thick aggressive blocks, shadows, negative space, etc.

And I got this:

Illustration: + my styleblock: no kanji. no hanko. silver rinpa school. use a felt-tip brush pen to ....

It's cool. I definitely got a goth female artist but I lost all the cool quotes. Apparently my style and quotes don't mix. Maybe because it's basically a graphic novel type style and you don't really see a lot of quotes in graphic novels.

I also tried it taking out the "silver rinpa school" but kept the rest the same and got this. I thought maybe the "silver rinpa school" might be interfering with the quotes but that wasn't the case.

Illustration: + my styleblock: no kanji. no hanko. silver rinpa school. use a felt-tip brush pen to ....

Image to Image Solution

So since I like my style but I wanted the quotes, it's easier to merge the two in image to image instead of typing out a bunch of quotes and hoping nothing gets mispelled.

So I went to image to image. You can use several different models but I chose Muse as I've been working with it a lot recently.

Doing Image to Image

To get to Image to Image,

  1. click on the pull-down menu on "Text to Image" button at top of the Create Image page

  2. click on "Image to Image"

  3. Illustration: Doing Image to Image

You can use one or more images as reference or start images for your image to image prompt. I used only one image. I used the guy as a start image in image 1.

Image 1: The image of the man shown above

My image to image prompt:

keep the setting. change the man in image 1 to solipsistic goth female artist dressed in lacepunk purple

change art style to 🖍️✒️🖍️. Keep the text from image 1. No additional text. no kanji. no hanko. silver rinpa school. use a felt-tip brush pen to ....

Notice I am saying to only change the man and the style but I want to keep the setting as is.

I added the bit about "No additional text. no kanji. no hanko" because "silver rinpa school" is a Japanese art movement which I like to use but sometimes kanji or a red stamp (a hanko) pop up unexpectedly. It's because the hanko is an artist signature but AI models have read them as part of the graphics. I don't want a hanko stamp on my goth image so "no hanko".

I can't say "no text" because I wanted the quotes so it was better to say "No additional text." This is useful if you're using a style that often results in text and you don't want to add text.

So my main intent was to do three things in the Image to Image

  1. 1. change the man to a goth female artist

  2. 2. change the art style

  3. 3. keep the quotes in the journal and on the wall

Final Image. This is the image I first showed you. I got my style, with a bit more color but that was ok. I kept all the quotes so I was very happy with that. and I changed the character to a female goth artist.

Illustration: So my main intent was to do three things in the Image to Image

Conclusion

That's how you change both the character and art style of an image but keep the setting. If you experiment with this technique, let me know. I'd love to see what you do.

Comments (30)

  • @CaylaCatz·

    I agree. I like that NC hasn't yet introduced a limit. If they do, I'll be in biiiig trouble, lol. Although to be fair, I wish I'd been doing this long time ago. I'm finding a couple of sites I used to use are gone so stuff there just disappeared. It's ok, they were never my main sites but still. We take for granted that stuff stays on sites but businesses disappear. Like when Warner bought Machinima (a bit video site where lots of people made videos in games like Blue vs Red.) Then AT&T bought Warner and within a year they closed down Machinima with no warning overnight. Thousands of people lost their videos, some of them years of videos that they never backed up on their own computers. Sad.

  • @CaylaCatz·

    I tried Word before and yeah, mine was a mess pretty early on. Then I tried notepad and saving the prompt with the picture in the same folder but I didn't keep up with that. Kind of a pain. This is kind of a pain too but I like it because I can actually find stuff now and I link to the pictures. I should try to do a tutorial on that. Not many people will do it but it might be useful for a few people anyways.

  • @CaylaCatz·

    thank you! Plus that is 5 images i can put in my database and erase from BP. I'm at my limit so right now, it's erase first, then create. I hated erasing tho as I was losing prompts. But I found a free database app so I feel much better because I can save my prompts plus everything is searchable. I wish I'd know before. It would be much easier to track stuff even if they are older.

  • @CaylaCatz·

    thank you! I could've typed in all the quotes but ... nah, lol. Plus I can use Muse in image to image and it's only 12 credits. Typing in that many quotes to generate a new image -- I'd probably go with GPT 2 medium at 60 credits. I might try Sunburst or Flare at medium 40 credits. It'd be good to test them to see how good they are at that much text.

  • @Sealine·

    Nice work! I definitely use image to image, sometimes to change the art style and sometimes to combine things I like from 2 different images. A recent example is the Freddy balloon video. I modified the balloon prompt but it resulted in a different girl, with an extra arm. Instead of trying again. I swapped the better balloon into the original image. GPT worked really well in that scenario.

Sign in to read all 30 comments and add yours.

More from @CaylaCatz

All posts →
On Style First: My Approach to Prompting

On Style First: My Approach to Prompting

I recently read @Cheinia's excellent post on upgrading AI portrait prompts: It's thoughtful, well-structured, and full of good advice. But one point made me pause—Step…

@justrob@justrob··2 min read
Side By Side Comparison

Side By Side Comparison

A few days ago, I commented on a "prompt enhancer" I use often. "Enhancer" might be the wrong word since it's also a prompt generator, but I feed old prompts into it and…

@paulie@paulie··1 min read

← All posts on the BudgetPixel blog