Otakubro Magic

'Magic' can effectively process scanned image (2nd) to restore them to a state closer to the original.

Just using nature language like:

Recently YouTube's recommendation algorithm pushed a 1996 arcade game, "Janken Game Acchi Muite Hoi! (Rock-Paper-Scissors, Look Over Here!/ジャンケンゲーム、あっち向いてホイ)", onto the homepage. One of the combat characters shown in the game's idle animation is a high school girl named tomomi, drawn by artist Akio Watanabe. As the character animation is very cute, it became a small hit on X as well.

Note: This so-called rock-paper-scissors game is where, after both sides show their choices, the winner shouts “Look over here!” while pointing in one of four directions—up, down, left, or right. If the loser turns their head to the same direction the winner pointed, they lose. It's basically an expanded version of a party game.

This time I grabbed onto the character's popularity and quickly made a few full-body images of JK characters as well.

So images 1–3 are the three different JKs I found from this search, and images 4–6 are pictures that more closely reflect the original game's style.

I used the following prompt this time:

Redraw this picture in TV animation art style, change to side view.

A way to generate different art style pictures of the same character (in different age).

Routine:

prompt: 'Draw a 18-years-old version of the girl in this picture, and change her pose to idle pose.' (3rd picture)

Make some picture contains two girls, another girl need to be in different artstyle. (Girl A(1st picture, left) is leaning over Girl B(2nd picture, right) , who is lying on her back, while Girl A props herself up on one elbow and gently caresses Girl B's neck with her other hand as they gaze at each other.) (4th picture)

prompt: 'Extract Girl A in this picture and change her pose into idle pose in front view.' (1st picture)

A simple idol-girl character from a Korean published RPG mobile game <Ascent Protocol>.

(Service period: July 31st to November 22nd, 2024)

For 1st and 2nd, it's some hires-processed images.

For 3rd-5th, it's some AI-generated full-body portrait based on 1st, 2nd, and original AI-generated (yes, the game really use AI for some character images used in main story😓) images.

For 6th-7th, just some motion-changed full-body portrait. And the last one is the original art style.

This time's I pick features fictional female jockeys from the horse-racing(?) game 'Winning Post 2 Final 97'. Besides having some independent storylines in the game, these two girls also have more outfits than other jockeys: one is a casual wear, and the other is a dress worn for celebratory parties like the Grand Slam.

5th picture shows the original in-game images of the two. The base resolution is roughly that size.

Japanese developer NTT Solmare (NTTソルマーレ) has an otome AVG mobile/web game brand series called "Shall we date?", which is fairly well-known among Japanese companies' English-market otome releases and covers a wide range of themes.

This was also the first time I discovered a serious Japanese mobile game company including yuri content in an otome game.

The first work I found was "Love Tangle", then I later found "Blood in Roses", and more recently discovered that in a version of "Lost Alice" included in the compilation app "Story Jar" there is a route for the Queen of Hearts — which is what this post will present.

As first two lines showed, Akari also seems to be based on the heroine of another title from the same company, "Guilty Alice".

This reminded me that there's a very similar Japanese otome called 白と黒のアリス but that one didn't include routes between "two alice", so I wouldn't make some AI-doujins for it…

Because "Story Jar" shut down in 2022, it's actually impossible to find the related CGs now — it's Lost Media.

Also, according to the walkthrough I seen from Amino-app, there should be one CG of the two of them on a date.

Since there's little chance of finding someone who played that version and grabbed the CG, I'll just make two myself.

In 2016, Ameba launched a female-oriented yuri visual novel titled "Kiss the Blooming Flower/花笑みにくちづけを" (Operating period: December 13, 2016 - May 12, 2022). Please note this should not be confused with another yuri-work titled "A Kiss for the Petals/その花びらにくちづけを" This title largely follows the gameplay model common among female-oriented browser games, focusing on outfit changes to accumulate points (attribute values) and balancing choices (story-oriented values) to unlock CG scenes. Developer R. INFINITY/アール・インフィニティ self-described it as "The industry's first 【G×G】! Female-oriented GL otome game" (Source: http://rinfinity.jp/games.php).

The game features three female LIs (though it seems to have dropped the teacher-student route).

This release features AI-enhanced CG of Megumi, the granddaughter of the academy's headmaster.

5th image depicts the Very Bad Ending CG; 6th image depicts the Happy Ending CG.

Enhance prompt is as usual:

Enlarge the picture in high definition, and make the color of the picture more vivid.

Original CG: https://tieba.baidu.com/p/5900884703

The CG descriptions for the Sylphiel (Sylphiel Belmore/シルフィール＝ベルモア) route in "Mahoutsukai to Goshujin-sama ~Wizard and the Master~(魔法使いとご主人様)" when generated using GPT, actually work quite well for creating cute girls light-yuri illustrations. Here's a reference for everyone：

[Place], two anime girls in a medium close-up: Girl A ([Characteristic of A]) faces the viewer with a shy smile and light blush, head tilted slightly left; her left hand touches the hair at her left temple while her right hand rests at her chest; Girl B ([Characteristic of B]) stands just behind and to A’s right, leaning in so her face is close to A’s right cheek with a warm, affectionate smile and blush; B wraps her arms around A from behind—left arm draped over A’s left shoulder with fingertips brushing A’s hair, right arm coming under A’s right arm and crossing A’s chest, the hands meeting near A’s sternum where A lightly holds B’s right hand; bodies overlap with A upright and B’s torso angled from right to left, B’s head slightly higher than A’s, conveying an intimate, tender mood.

Another one use descriptions from Lacey route @ <Corona Borealis>:

[Place], Girl A ([Characteristic of A]) stands left-center facing slightly right, shoulders tucked and head tilted right with one eye squeezed shut, a small blush and a sweat drop on her cheek; she grips her left shoulder strap with one hand and clutches her opposite wrist with the other; Girl B ([Characteristic of B]) is positioned behind and to the right of Girl A, leaning in from the right, kissing Girl A’s cheek just in front of the ear, her arms wrapped around Girl A’s upper chest and neck—one hand cupping Girl A’s jaw/neck and the other resting across her collarbone; camera is a close, waist-up three-quarter view with Girl B’s hair flowing across the scene.

It's also very easy to create a kiss illustration...lol (like the 3rd one) but its prompt is far more eaiser:

Girl in white kisses Girl in black on her lip

Finally, here is the prompt to let ChatGPT to generate description for CG image.

Using as a universal template (girl A, girl B, etc.), describe all the character's action details and positional details in the image with a single English prompt. No need to include clothing or background details.

I mainly created some illustrations for several girls from the kenka-banchou series.

I made them in the same routine: grab a set of screenshots with 3D model images, then use the prompt "HD restoration + draw something like this: she is in full-body…"

Yep, even 3D models from the PSP era can be restored into beautiful full-body illustrations—it just requires taking a few more screenshots.

Then it's not even finished yet—there are still three or four girls who should be able to get their character art done.

The four works above were achieved using several techniques:

Repeatedly referencing the original line art to color the artwork,

Filling in areas with incorrect details to modify them based on the correct artwork,

Creating a full-body character illustration from this angle by referencing other perspectives of the character, and the simplest two methods:

Modifying actions and expressions through natural language.

HD restoration.

*The second image was colored using line art techniques only without far more adjustment.

Using 'HD restoration of this image in a large scale' to enhance and restore low-resolution magazine images. Since the model excels at alphanumeric characters, it may require minor adjustments for Japanese kanji, hiragana, and katakana.