Demo : 當今最紅女團 @ 港鐵站

[mtr]

https://www.instagram.com/p/DaF73lGhNo3/

[mtr]

Demo : 當今最紅女團 @ 連登 [18+]

https://lihkg.com/thread/4124129/page/3

https://na.cx/i/yH1xQhL.gif
https://na.cx/i/WUub5TK.gif
https://na.cx/i/ws7pGs0.gif

[mtr]

本地二創推介

https://www.threads.com/@michael.yung

[joe]

[mtr]

有人問我點先可以整到佢地四位各有自己造型風格，Gen AI 最緊要畀 guideline 同boundary 佢，個 base 最好唔要其他物件影響
所以人物設定圖好緊要，最好仲要係冇任何服飾。但主流Gen AI 唔畀裸露嘅時候，一件素色打低就做足夠
Gen 圖時，加埋人物設定去畀 AI 參考，出嚟嘅效果一定唔會差得去邊。
分享我用嘅人物設定圖畀大家
有草草、雪兒、小悠、可樂
希望大家用得著

https://www.threads.com/@michael.yung/post/DaKGkPvgS7_

[mtr]

想知本身點樣先出到呢一張設定圖

先 capture 幾張 video 嘅人物畀 chatgpt, 叫佢整文字版 factsheet, 你睇完佢對呢個角色嘅理解係你想要嘅嘢後，就叫佢切人物造型圖
如果你係諗住用其實服飾，就指明唔要 screenshot 入面嘅服飾，只需要內搭（啫係打底）畀你之後套其他衫落去

[mtr]

[18+]

https://lihkg.com/thread/4124129/page/9

https://na.cx/i/tDLnznQ.gif
https://na.cx/i/gC4C7HV.gif
見無人答 我用this
https://t.me/PleasureLabAIbot

Demo : 當今最紅女團 @ 連登 [18+]

https://lihkg.com/thread/4124129/page/3

https://na.cx/i/yH1xQhL.gif
https://na.cx/i/WUub5TK.gif
https://na.cx/i/ws7pGs0.gif

[joe]

[mtr]

香港經典校服 look，"可樂" 最正。留言有新法書院校服，有埋 St Paul Convent 的就更好。

https://www.threads.com/@obsessions_fan ... PGRYQE1mY/

[joe]

[mtr]

有埋 St Paul Convent 的就更好。

初戀就是聖保祿女生，無聯絡多年，已經唔記得 St Paul Convent 對校服鞋係唔係附圖個款，唔係用 ComfyUI 慢慢根據樓上位大大提供的人物造型圖設定砌，只是用 ChatGPT image2，希望無 distort 左 "草草"個樣。

Pret-A-Manger 係小弟同初戀第一次撐檯腳既食肆，就生成這張圖留個記念吧。

[joe]

[mtr]

Further reading : how to create/write prompts for a character sheet like the one below :

https://x.com/tonomisugi/status/2073218303452123318

[mtr]

A prompt for generating a character sheet.





https://x.com/A_I_V_Y/status/2073527631694434349

Create a character sheet using the person/character in the attached image as the primary reference, ensuring it depicts the exact same person/character.

If the attached image is a photograph, maintain a photorealistic style; if it's an illustration, anime, manga, 3D, or deformed/caricature style, preserve that exact art style, linework, coloring, texture, and level of deformation.
Do not arbitrarily convert it to photorealistic, illustration, or a different art style.

The purpose is not for profile introductions but as visual reference material for character creation, AI image generation, and model consistency.
Do not include profile elements such as name, age, personality, hobbies, or descriptive text.
For text within the image, include only short Japanese labels indicating each panel. Do not use English notation.

Image format:
A single wide image in 16:9 aspect ratio. Background in white to light gray. High resolution. A clean, easy-to-read character sheet layout organized with thin ruled lines or boxes.

Primary conditions:
Faithfully reproduce the face, eyes, eyebrows, nose, mouth, contours, hairstyle, hair color, skin and coloring texture, body type, overall atmosphere, and clothing impression from the original image.
Do not make it a different person or character.
Prohibit excessive beautification, simplification, body type changes, or face redesigns.
Exclude temporary small items, food, backgrounds, poses, or handheld items from the original image unless necessary for the character design.

Clothing:
If the clothing in the original image is clear, maintain it.
If the full body is not visible, supplement with natural clothing that fits the atmosphere of the original image.
For front, side, and back views, keep the same clothing, same hairstyle, and same body type consistent.
Avoid excessive exposure, underwear, swimsuits, or sexualized outfits.

Content to include:
[Full body] Front, side, back
[Face close-up] Front, profile, 45-degree angle
[Expressions] Neutral, smile, beaming smile, serious, surprised, embarrassed, thinking, troubled
[Face parts] Eyes, eyebrows, nose, mouth, ears, contours, skin, texture
[Hair details] Bangs, side hair, back hair, hair flow
[Other angles] Diagonal left, diagonal right, from above, from below, back of head

Layout:
Make it easy to read as reference material, with clean margins and each item organized in boxes.
Do not use magazine profile style, resume style, or poster style.
Do not include long explanatory text; use only short Japanese labels.

Negative specifications:
Different person, different character, changed face, changed art style, arbitrary photorealistic conversion, arbitrary anime conversion, hairstyle change, hair color change, inconsistent clothing, body type change, excessive beautification, excessive simplification, profile text, name, age, personality, hobbies, English labels, long explanatory text, garbled text, hard-to-read text, sloppy layout, low resolution, distorted face, weird eyes, hand errors, finger errors, duplicated faces, sexualized outfits, underwear, swimsuits, unnecessary small items, holding food in mouth, random objects, cluttered background

[mtr]

Here's how to use it—it's simple.

① Upload one image you want to base it on
② Paste the prompt above
③ Use the generated character sheet as a reference for future attempts

That's all.

If you want to keep characters consistent in AI image generation,
it's recommended to start by creating a character sheet.