This post is about helping @[email protected] in creating a Perchance style for Virtual Tabletop tokens. (From this Perchance forum post)
Generator used is x2x’s text2image-generator
Here are my prompts and images in my current testing:
Japanese Token
style: No Style
prompt: (3d model in top down view:1.1), (from above:1.3), from behind, plain background, person with a katana, edo style
n_prompt: soft shadows, hard shadows, reflections
res: 512x512
seed: random
Thief in white Clothing
style: No Style
prompt: (3d model in top down view:1.1), (from above:1.3), from behind, plain background BREAK a male thief with white clothes knife in hand, fantasy d&d style
n_prompt: soft shadows, hard shadows, reflections, disc
res: 512x512
seed: random
We can probably increase more emphasis on the from above or at the 3d model in top view, but any higher than 1.5 is muddy.
Essentially adding:
(3d model in top down view:1.1), (from above:1.3), from behind, plain background
Before your actual prompt will make the image like it is from top down.
Here’s what I managed so far. I generated alot of people that were viewed from above, yet the person was looking upward. Easy to do landscape photography and vehicles and even a cupcake, lol. These are the only people I managed to do correctly (tho the person is distorted), which hints that it IS possible, but I’m not sure how yet. something like ((google maps image)) apparently does good overhead landscape photos. (from above) sometimes is pretty effective. and at least one of these very rare actual top down person images was from a prompt about ‘drone photography’. currently the style on beautiful-people I have doing this mix and matches from a small list that I am refining to this very day!