Saturday, October 03, 2026

AI-synthetic: Sana Myon Sanae & Myon Youmu Touhou Project

 

https://www.pixiv.net/en/artworks/150428883 exhibits the AI-synthetic illustrations of Touhou 13 ~ Ten Desires generated by ChatGPT plus processing my own drawing. I am shifting back from this AI-synthetic experiment to hand-drawing art work. 

Original image / composition reference:
https://www.pixiv.net/en/artworks/121630821

Related original post / scene reference:
https://art-blue-liberalism.blogspot.com/2024/08/sana-myon-sanae-myon-youmu-touhou.html

━━━━━━━━━━━━━━━━━━━━

Redraw Request Prompt

Redraw the attached base illustration while preserving the overall composition, character placement, interaction, close paired arrangement, central hand-contact motif, camera framing, and general emotional tone as closely as possible.

Use the attached base illustration as the PRIMARY VISUAL REFERENCE for:

  • overall composition
  • left-right character placement
  • close distance between the two characters
  • general pose structure
  • central hand interaction
  • upper-body framing
  • character scale
  • paired portrait feeling

Use the second reference image as an expression and scene-context reference, especially for the left character’s slightly serious, mildly frustrated, faintly exasperated, and subtly incredulous emotional nuance, while keeping the final expression graceful and aesthetically harmonious within the paired portrait.

Use the third reference image as a skin tone and skin-rendering reference, especially for:

  • soft luminous complexion
  • smooth skin texture
  • delicate rosy blush
  • creamy anime-style tonal transitions
  • subtle soft highlights
  • refined dimensional skin shading

Use the additional reference image of the pink-haired ghostly noblewoman only as inspiration for the right character’s facial elegance, warm illumination, delicate skin rendering, and the treatment of surrounding ghostly wisps, while preserving the right character’s actual identity, hairstyle colour, costume identity, placement, and pose.

Use the small cropped reference of the left character’s original expression and torso area as an additional guide for her intended mood and pose energy, but prioritize:

  • anatomical coherence
  • correct shoulder orientation
  • correct arm structure
  • correct hand placement
  • natural torso construction

over any accidental distortions present in a flawed previous version.


Overall Art Direction

Redraw the illustration as a polished early-to-mid-1990s Japanese fantasy RPG key illustration, reflecting the elegant visual language of Japanese fantasy strategy-game promotional art, anime illustration books, and romantic shōjo-fantasy artwork from approximately 1991–1995.

The final image should combine:

  • the dramatic clarity and heroic presence of early-1990s Japanese RPG promotional art
  • graceful character silhouettes
  • romantic fantasy colour harmony
  • jewel-like anime eyes
  • decorative but readable costume design
  • elegant supernatural atmosphere
  • softly luminous skin rendering
  • refined linework
  • richly separated hair strands
  • soft painterly cel shading
  • atmospheric background painting

Preserve the core composition and relationship of the base image, while significantly refining:

  • anatomy
  • hand construction
  • face structure
  • line quality
  • costume accuracy
  • lighting
  • shading
  • skin rendering
  • background atmosphere
  • supernatural ambiance

Wednesday, September 30, 2026

AI synthetic: Sanae Kochiya vs Kogasa Tatara - Touhou Project Fan-art

 

The image generation of ChatGPT started freezing while my repeated attempts to correct Kogasa’s costume continued to produce deviations from the intended design as due to some ongoing issue of ChatGPT on 30th Sept. 2026. So, I have decided to use this version an illustration of Kogasa Vs. Sanae as a compromise for now.

AI Redraw Request — Crying Umbrella Youkai Defending Against a Divine Shrine-Maiden 

Saturday, September 12, 2026

The trouble some issues encountered while modelling the economic growh together with a new exponential modelling

 

The following text is AI (ChatGPT plus) generated with reference to my current issue in mathematical modelling of the macroeconomic growth theory. 

---  

This week I had an interesting little incident at the boundary between economics and mathematics.

I was checking an exponential model for an economic growth / Total Factor Productivity (TFP) problem. After going back to the mathematical formulation and implementing it carefully in MATLAB, the objective function Q(ϕ)Q(\phi) showed a clear minimum.

At first sight, that sounds encouraging: the optimisation problem has a well-defined solution.

However, when I reconstructed the corresponding TFP trajectory using the parameters associated with that minimum, the result did not reproduce the observed trajectory satisfactorily.

So I found myself in an interesting situation:

the mathematics said, “Here is the optimum,” while the economic data replied, “Not so fast.” 😅

This does not necessarily mean that the exponential approach is mathematically wrong. Rather, it raises a more interesting question: is this particular model structure appropriate for representing the empirical behaviour of TFP?

At the moment, I am therefore also comparing the result with a penalised smoothing spline, which appears to represent the trajectory more naturally.

For me, this has been a useful reminder that finding a mathematically neat optimum is not the same thing as finding a good empirical model. Optimisation, model specification, and interpretation all have to work together.

The investigation is still ongoing, so this is not a final conclusion — just one of those small research episodes where economics and mathematics refuse to cooperate quite as politely as expected.