Resized to 47% of original (view original)
| Prompt | Same art style, close-up upper body direct top-down view of very busty woman wearing strapless dress, lying on her side on a couch (oriented vertical in relation to the image), arms up (both sides of head). Smug expression, sideways glance toward viewer. She wears a gold necklace and a black wrist band and thin gold bracelet on one wrist (see reference). |
|---|---|
| Negative prompt | Unknown |
| Model | gemini-3.1-flash-image-preview and gemini-3-pro-image-preview |
| Note | Additional work between prompts, and clean-up after second prompt. |
| Prompt 2 | 3:4 aspect ratio, same art style as <image_0>. <image_1> is reference only. Close-up top-down vertical view of woman lying on her side on a couch, smug grin expression, sideways glance at viewer. POV containing viewer's arms, wearing black military jacket, left hand playfully grabbing her wrist, and right hand grabbing top of her head. Her hand on that side holds viewer's wrist. She only wears bracelets on one wrist (left side of image). |
| Prompt 3 | 3:4 aspect ratio. Reconstruct the composition of <image_0> exactly as given, in the art style of <image_1>, without image compression artifacts. A top-down view of a woman lying on her side on a brown couch with her hands up. She glances sideways at the viewer with a smug smirk. POV from the viewer who wears a black commander jacket; his left hand grabs the woman's wrist, and his right hand tightly grabs the woman's head. |

Danbooru post #11331248 as input. Inspired by Danbooru post #3644796 but was not used as input due to filter issues with corpo models without obscuring most of the image. Actually I did get something through NB but it wasn't looking good, so I did it "from scratch".
It was near impossible to obtain desired results from prompting alone, so I broke it up into steps. At first I genned some without the POV hands, then added POV hands. I was going to be finished earlier but noticed the arms coming from the bottom of the image just looked wrong, then I moved them to the sides.
After iterative deepfrying with NB2, I switched to NBP with the "reconstruct" prompt to get a larger, clean version.
I wanted to try it with gpt-image-2 but it filters my NB outputs that I was going to use as reference, so this wasn't going to happen.
The idea for this came after post #160714's filename.