Resized to 55% of original (view original)
| Prompt | 3:2 landscape image, historical painting. This woman squats, knees not touching ground, to face three Moogle lined up side by side. The woman wears a short blue dress and single fishnet thigh-high, and has a flat circle behind her head to represent her sainthood. The first moogle presents an URGI AR-15 by holding it sideways with both hands. The second moogle holds a pair of magazines (right side of third image). The third moogle holds a fragmentation grenade (middle of third image). |
|---|---|
| Negative prompt | Unknown |
| Model | chatgpt-image-latest (20251216) and gemini-3-pro-image-preview |
| Note | Prompt 2 didn't change the inserted mag, and I decided to replace the rifle entirely due to odd parts, hence Prompt 3. As always, I use an image editor to discard unedited parts otherwise additional quality loss occurs on each prompt. |
| Prompt 2 | 3:2 image. Fix image_2 so the middle moogle is holding two magazines matching image_0, and the magazine inserted in the rifle is correct. I have circled the parts requiring correction. |
| Prompt 3 | 3:2 image. The rifle in image_1 should match image_0. |
| Prompt 4 | The woman in image_1 is squatting. Ensure that her hands are in logical positions such as on top of her legs. She wears black gloves. Retain painting style. |
Image input: crop of Danbooru post #5306179, asset #583641, asset #583640
GPT for base image generation with best looking Moogles.
NBP to fix gun, magazines, and hands (generally more accurate at everything, disregarding style).
NAI for a little bit of inpainting, also reminded me that the left earring is actually white, which is barely peeking in the reference.
asset #583643: Bonus text generated with gemini-3-pro, 3 lines edited. I apologize for the slop.
I really wanted the stock to be FDE colored, but I accidentally selected 1024x1023 during my previous generations, and I hit lmarena rate limit.