img2img for left: asset #521839 It's screwy trying to do more one part at once, so I did these separately. The model kept putting her close to the middle, so I found a trick to move her over by telling it to draw a red square in the middle first and don't let anything touch it.
top right: post #134891 Does it finally work with NAI's CR?.. No. All this work and I'm being gatekept so hard from just making posts since I didn't want to be an AI off-model slopper. Ignore the cleavage hole and ID card, then this character would be the easiest thing to prompt.
img2img for bottom right: asset #521843 The model copied my doodle of the logo from the reference, so I went into an image editor and used shape tools to redraw it. For the portrait, the model has zero clue about closing the top button, usually not generating the button at all. If you make it add a button, it's not going to know how to "close it". Had to inpaint.
Prompt
My input is reference only; do not copy and paste from the reference. First, draw a small red square in the center of the image. You will not allow anything to touch this. On the left side, generate a three-quarter view of this woman from above the knees, on plain white background with no space to the left side. She has cyan eyes and long brown hair tied into a high ponytail with white hair scrunchie. She wears a black pencil miniskirt and a sleeveless, strapless off-shoulder shirt with large frills for the neckline. The neckline is extra low, exposing significant cleavage above it. Most notably the center top of frills contains a circular hole tied at the top by a white ribbon string. The hole is is a perfect circle as if an o-ring. In the entire upper right quadrant, zoom in by 4x and create a heavily enlarged inset of the hole and string ribbon without the lanyard. DO NOT COPY THE REFERENCE IMAGE. In the main image, not the inset, she should wear a thin blue lanyard with a vertical ID card with an empty portrait in the middle. The only text below the portrait should read "천소영" on line 1, and "마케팅 기획부" on line 2. The company logo is shaped like a 4 inside a circle. The ID card holder is a solid blue stripe across the top, and blue triangle bottom corners. The lanyard goes over the shirt and should not disappear through the hole. Leave the bottom right quadrant of the image empty.
Negative prompt
Unknown
Prompt 2
Generate a clean digital illustration of the ID card, leaving the portrait blank. Assume this is one solid card instead of a badge holder. The top is colored blue. The company logo is shaped like a simplified 4 touching the inside of the circle. Next to the logo is a black line. There is a black line below the portrait. The bottom corners are blue. The text below the portrait should read "천소영" on line 1, and "마케팅 기획부" on line 2. There is another black line below the text. The second line of text and the black line below it crosses into the blue corners. Don't make the card too big, it should barely fit within one quadrant of the full image.
Prompt 3
Generate a portrait of this woman for the ID card. In this front view portrait, she wears a fully buttoned white dress shirt and black suit jacket, and is smiling. Solid pale light blue background. Don't spread her ponytail. We should barely be able to see her sidelocks against the background.
img2img for left: asset #521839
It's screwy trying to do more one part at once, so I did these separately. The model kept putting her close to the middle, so I found a trick to move her over by telling it to draw a red square in the middle first and don't let anything touch it.
top right: post #134891
Does it finally work with NAI's CR?.. No. All this work and I'm being gatekept so hard from just making posts since I didn't want to be an AI off-model slopper. Ignore the cleavage hole and ID card, then this character would be the easiest thing to prompt.
img2img for bottom right: asset #521843
The model copied my doodle of the logo from the reference, so I went into an image editor and used shape tools to redraw it.
For the portrait, the model has zero clue about closing the top button, usually not generating the button at all. If you make it add a button, it's not going to know how to "close it". Had to inpaint.