Cast designs: shortlist README, shared style rules, prior-art facts
Seven characters, Persona-style chibi bust portraits, render process with one hypothesis per prompt. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Meta Vens
parent
8169ee5cd6
commit
352da2765b
2 changed files with 217 additions and 0 deletions
126
cast-designs/README.md
Normal file
126
cast-designs/README.md
Normal file
|
|
@ -0,0 +1,126 @@
|
|||
# Cast designs: the shortlist
|
||||
|
||||
Seven characters get real designs before anyone else in the swarm does.
|
||||
The operator, 2026-10-10: "none of these guys are gonna make it to the swarm.
|
||||
I want Libby, Heimdall, Kiko, Aqua, Weksa, Nibu, Eureka."
|
||||
|
||||
| Character | Body (repo) | Key colour | File |
|
||||
|---|---|---|---|
|
||||
| Eureka | the Mind herself | ember orange `#ff8a2a` | [eureka.md](eureka.md) |
|
||||
| Libby | CultLib | reading-lamp green `#2fbf71` | [libby.md](libby.md) |
|
||||
| Heimdall | Heimdall | cobalt `#3466ff` | [heimdall.md](heimdall.md) |
|
||||
| Kiko | StreamPixels | stream magenta `#ff3fa4` | [kiko.md](kiko.md) |
|
||||
| Aqua | AquaSynth | tank cyan `#22d3e6` | [aqua.md](aqua.md) |
|
||||
| Weksa | weksa | terracotta `#d9483b` | [weksa.md](weksa.md) |
|
||||
| Nibu | AetheriaLore | violet `#9a6bff` | [nibu.md](nibu.md) |
|
||||
|
||||
Each file holds three design directions (A, B, C) that differ in silhouette
|
||||
and concept, a ready-to-run design-sheet prompt per direction, and the
|
||||
portrait set spec. Facts behind the style rules are in
|
||||
[research/style-prior-art.md](research/style-prior-art.md).
|
||||
|
||||
## What failed, so it is not repeated
|
||||
|
||||
The sticker rig (`self/cast-stickers`) drew every character on one shared body
|
||||
and swapped hair and props. At thumbnail size all 24 read as the same figure in
|
||||
different wigs: tiny full bodies, no dominant shape, no hook. These designs
|
||||
start from the opposite end: silhouette first, one hook per character, and the
|
||||
face large enough to carry the expression.
|
||||
|
||||
## Style target
|
||||
|
||||
Chibi anime **bust portraits** in the manner of Persona dialogue UI: head and
|
||||
shoulders, angled toward the text box, one expression per line of dialogue,
|
||||
inside a graphic frame keyed to the character's colour. Persona Q is the
|
||||
series' own precedent for this: Soejima redrew the P3/P4 casts as
|
||||
super-deformed and kept each character's standout features exaggerated so
|
||||
they survived the shrink.
|
||||
|
||||
- Proportions: 2.5 heads tall on the full-body sheet (the 2 to 3 head chibi
|
||||
band). In the bust, the head fills 55 to 65 percent of frame height.
|
||||
- The bust is the product. The full-body turnaround exists to lock the design;
|
||||
it is reference, not a sticker.
|
||||
|
||||
## Shared rules: what makes the set cohere
|
||||
|
||||
The set coheres by **rendering**; the characters differ by **shape and hue**.
|
||||
|
||||
1. **Line.** Bold dark outline, outer silhouette line about twice the weight of
|
||||
inner lines. Line colour is a dark shade of the character's key colour, never
|
||||
pure black.
|
||||
2. **Rendering.** Flat cel shading: one base tone and one shadow tone per
|
||||
material, plus a single rim light in the key colour on the side away from
|
||||
the light. No gradients on skin, no texture, no painterly blending, no pixel
|
||||
art, no 3D look.
|
||||
3. **Eyes.** Large anime eyes with two specular highlights, same construction
|
||||
for everyone. Eye shape (round, almond, sharp, hooded) is a per-character
|
||||
choice and part of the design.
|
||||
4. **Framing.** Bust cropped at mid-chest, three-quarter view, character faces
|
||||
screen right toward the text box; light from upper left. The baseline (shoulder
|
||||
line, eye line) stays fixed across a character's whole expression set so the
|
||||
portrait does not jump between lines.
|
||||
5. **Palette discipline.** Each character owns one key hue, fixed across all
|
||||
three directions, and no two characters share it. A direction's palette is 4
|
||||
to 6 colours: key, a dark of the key for line and deep shadow, one skin or
|
||||
body tone, one neutral, one accent. The accent may borrow another
|
||||
character's hue only if it covers under 10 percent of the figure.
|
||||
6. **UI frame.** Portraits sit on the GameCult ground `#07111a` in a
|
||||
parallelogram card, edge-lit in the character's key colour, nameplate in IBM
|
||||
Plex Mono uppercase tracked wide (the brand label face). The frame is
|
||||
composited, never rendered into the image.
|
||||
|
||||
## What keeps them from being samey
|
||||
|
||||
- **One dominant shape language per character**, and each one different:
|
||||
Libby squares, Heimdall arcs and spikes, Kiko stepped pixels, Aqua circles,
|
||||
Weksa flowing lines, Nibu swept wedges, Eureka a spark on a circle.
|
||||
- **The hook is above the shoulders.** A bust only shows head and shoulders, so
|
||||
every hook is hair, headwear, ears, a held prop raised to the face, or a
|
||||
collar. A hook on the boots does not exist in this product.
|
||||
- **Silhouette test, cast-wide.** Fill each bust black, shrink to 48 px, put all
|
||||
seven side by side. Every one must be nameable. If two collide, the later
|
||||
direction changes, not the rendering rules.
|
||||
- **No shared body.** Body type, head shape, eye shape and hair mass are
|
||||
per-character design decisions written into each direction.
|
||||
|
||||
## Render process
|
||||
|
||||
From `~/.claude/doctrine/images.md` and the concept-mill rule
|
||||
(image-prompts-test-hypotheses):
|
||||
|
||||
1. **Every render tests one stated hypothesis.** Each prompt below states it.
|
||||
Record hypothesis, variable, observation and the operator's verdict in a
|
||||
results table beside the renders.
|
||||
2. **One variable per variant, against a named baseline.** The first render of
|
||||
a direction is its baseline. A follow-up changes one thing (hook size, palette
|
||||
accent, the style block's line clause) and says which.
|
||||
3. **The exact prompt is saved beside each image**, same base filename, `.md`
|
||||
with the generation mode or model recorded. An image without its prompt is
|
||||
not a deliverable.
|
||||
4. **Renders are from scratch.** Carry what worked forward by rewriting the
|
||||
prompt, never by editing a prior render or feeding it back as a reference.
|
||||
5. **The style block is not sacred.** When a round points at a clause, edit it,
|
||||
name the edit as that round's variable, and promote what holds back into this
|
||||
README.
|
||||
|
||||
Suggested layout: `cast-designs/renders/<character>/<direction>-r<round>-<variant>.png`
|
||||
with its `.md` prompt twin, and `cast-designs/renders/<character>/results.md`.
|
||||
|
||||
### Style block v1 (inlined in every prompt)
|
||||
|
||||
> chibi anime character design, 2.5 heads tall super-deformed proportions,
|
||||
> oversized head, large expressive anime eyes with two highlights, bold clean
|
||||
> outline thicker on the outer silhouette than on inner lines, line colour a dark
|
||||
> shade of the key colour, flat cel shading with one shadow tone and one rim
|
||||
> light, no gradients, no texture, no pixel art, no 3D render. Character design
|
||||
> sheet on a plain off-white background: on the left a large bust portrait, head
|
||||
> and shoulders cropped at mid-chest, three-quarter view facing right; on the
|
||||
> right a full-body chibi turnaround, front, three-quarter, side and back, neutral
|
||||
> standing pose, all at the same height. No text, no logos, no watermark.
|
||||
|
||||
## Order of rendering
|
||||
|
||||
Round 1 renders one direction per character, chosen for the strongest
|
||||
above-the-shoulder hook, to test the style block itself across the cast before
|
||||
spending renders on alternatives. See the end of each character file for its
|
||||
candidate and the reason.
|
||||
91
cast-designs/research/style-prior-art.md
Normal file
91
cast-designs/research/style-prior-art.md
Normal file
|
|
@ -0,0 +1,91 @@
|
|||
# Style prior art: Persona portraits, chibi proportions, iconic reads
|
||||
|
||||
Facts gathered 2026-10-10 for the cast-designs shortlist. Each line says what the
|
||||
source states; where no primary source was found, the line says so. Conclusions
|
||||
drawn from these facts live in `../README.md`, not here.
|
||||
|
||||
## Persona series art direction and UI
|
||||
|
||||
- Shigenori Soejima has been the Persona series character designer and art
|
||||
director since Persona 3 (2006), succeeding Kazuma Kaneko.
|
||||
Source: [Wikipedia, Shigenori Soejima](https://en.wikipedia.org/wiki/Shigenori_Soejima).
|
||||
- Each mainline game is keyed to one colour: art director Masayoshi Sudou says
|
||||
Persona 3's main colour was teal, Persona 4's "a vivid yellow", Persona 5's
|
||||
"a crimson red", with black and white text chosen against it.
|
||||
Source: [Famitsu interview translated by SegaBits](https://segabits.com/?p=54918).
|
||||
- Kazuhisa Wada says Persona 5's UI design and red-and-black colouring were chosen
|
||||
from the game's theme and message, and contrasts P5's warm colours with
|
||||
Persona 3 Reload's cold ones.
|
||||
Source: [GamerBraves interview](https://gamerbraves.com/?p=129988).
|
||||
- Sudou and Soejima met often during Persona 3 to brainstorm the GUI; the same
|
||||
look extended to the website.
|
||||
Source: [SegaBits / Famitsu](https://segabits.com/?p=54918).
|
||||
- Persona 5's interface makes heavy use of character portraits and 3D models;
|
||||
Joker stays in the menu background and changes pose with the selected item.
|
||||
Source: [Mechanics of Magic, Visual Design of Games: Persona 5](https://mechanicsofmagic.com/2022/04/23/visual-design-of-games-persona-5/).
|
||||
- Persona 3 Reload remade the dialogue portraits and kept characters' signature
|
||||
expressions; one cited example is the protagonist's deadpan stare.
|
||||
Source: [Yahoo / GamesRadar on Reload portraits](https://sg.news.yahoo.com/persona-3-fans-really-hope-162644075.html),
|
||||
[Gematsu, Reload announcement](https://www.gematsu.com/2023/06/persona-3-reload-announced-for-xbox-series-xbox-one-and-pc).
|
||||
- Not found in a primary source: the exact expression count per character, or a
|
||||
written spec for dialogue-portrait framing. What the README says about bust
|
||||
framing (head and shoulders, angled toward the text box, one expression per
|
||||
line of dialogue) is observed convention across the games, not a cited rule.
|
||||
|
||||
## Persona Q: the series' own chibi redesign
|
||||
|
||||
- Persona Q (3DS, 2014) redrew the P3 and P4 casts as super-deformed chibi; Soejima
|
||||
did the redesigns and said it was his first time working in that style.
|
||||
Source: [Wikipedia, Persona Q](https://en.wikipedia.org/wiki/Persona_Q:_Shadow_of_the_Labyrinth),
|
||||
[Game Informer feature, 2014](https://www.gameinformer.com/b/features/archive/2014/07/18/a-glimpse-into-the-mind-of-personas-art-director.aspx).
|
||||
- His two stated reasons: kinship with Etrian Odyssey's look, and the small 3DS
|
||||
screen, where full-proportion designs would look cramped.
|
||||
Source: [Nintendo Everything](https://nintendoeverything.com/tag/shigenori-soejima/),
|
||||
[Nintenderos (Spanish)](https://www.nintenderos.com/2014/10/el-disenador-de-persona-q-nos-habla-sobre-el-estilo-del-juego/).
|
||||
- His method, as reported: find what makes each character stand out, then adjust
|
||||
those features so they stay recognisable after deformation.
|
||||
Source: [Game Informer](https://www.gameinformer.com/b/features/archive/2014/07/18/a-glimpse-into-the-mind-of-personas-art-director.aspx).
|
||||
|
||||
## Chibi proportions
|
||||
|
||||
- Chibi / super-deformed figures are usually 2 to 3 heads tall; guides range up
|
||||
to 4, and SD is quoted at 2 to 5 depending on style and age.
|
||||
Sources: [Wikipedia, Chibi (style)](https://www.wikipedia.org/wiki/Q-version),
|
||||
[Clip Studio tips, chibi body ratio](https://tips.clip-studio.com/es-es/articles/4806?org=1),
|
||||
[intfiction forum thread](https://intfiction.org/t/video-games-of-characters-with-short-proportions/63241).
|
||||
- Wikipedia puts the SD head at one third to one half of total height and notes
|
||||
that details such as jacket folds are dropped in favour of general shapes.
|
||||
Source: [Wikipedia, Chibi (style)](https://www.wikipedia.org/wiki/Q-version).
|
||||
- Terminology is loose: "SD" names the head-to-body ratio, "chibi" connotes small,
|
||||
round and childlike; the terms are used interchangeably. Source: forum
|
||||
discussion above (opinion, not a reference work).
|
||||
|
||||
## Iconic reads at small size
|
||||
|
||||
- Valve's TF2 paper (Mitchell, Francke, Eng, NPAR 2007): the nine classes were
|
||||
given distinct silhouettes and remain identifiable as pure silhouettes; the
|
||||
silhouette-only test was used during concept work.
|
||||
Source: [Illustrative Rendering in Team Fortress 2](https://www.cs.princeton.edu/courses/archive/fall07/cos597B/papers/mitchell-team-fortress.pdf).
|
||||
- The same paper: body proportions, weapons, and lines made by hats, footwear and
|
||||
clothing folds were designed to make each outline unique; interior folds echo
|
||||
the outer silhouette.
|
||||
Source: same paper; [Valve NPAR07 slides](https://cdn.fastly.steamstatic.com/apps/valve/2007/NPAR07_IllustrativeRenderingInTeamFortress2_Slides.pdf).
|
||||
- TF2's read hierarchy: team from colour, class from silhouette and shape,
|
||||
weapon from the rest of the outline.
|
||||
Source: [GDC 2008, Stylization With a Purpose](https://cdn.steamstatic.com/apps/valve/2008/GDC2008_StylizationWithAPurpose_TF2.pdf).
|
||||
- Practitioner guides (blog-grade, rules of thumb): fill the figure black and view
|
||||
it at about 32x32 px; check the whole cast side by side; circles read friendly,
|
||||
squares stable, triangles sharp or dangerous; keep 2 to 5 core colours with one
|
||||
signature accent; order of work is silhouette, then colour, then detail.
|
||||
Sources: [Bugnet, character design basics](https://bugnet.io/blog/character-design-basics-for-indie-devs),
|
||||
[Made Good Designs](https://madegooddesigns.com/character-design-basics/).
|
||||
|
||||
## Expression sets in portrait-driven games
|
||||
|
||||
- Commercial visual-novel sprite packs ship 6 to 33 expressions per character;
|
||||
a common core is neutral, smile, sad, angry, surprised, often plus one
|
||||
character-specific state. Packs keep framing and baseline identical across
|
||||
expressions so the sprite does not jump when it changes.
|
||||
Sources: [itch.io Luna collection](https://prometheus-pictures.itch.io/luna-emotional-sprite-collection-33-expressions-for-visual-novels-rpgs-romance),
|
||||
[itch.io convenience-store worker set](https://pompack.itch.io/friendly-convenience-store-worker).
|
||||
Market practice, not an industry standard.
|
||||
Loading…
Reference in a new issue