docs(ui): lock Japanese battle direction

This commit is contained in:
2026-08-12 01:06:47 +08:00
parent b13a026e52
commit b8e72f0ead
14 changed files with 432 additions and 0 deletions
@@ -0,0 +1,29 @@
# Phase 6 UI Direction Generation Attempts
## Attempt 01 — `INVALID_NETWORK_ENVIRONMENT`
- UTC: `2026-08-11T16:30:26Z`
- Provider/model: API Mart / `gpt-image-2`
- Intended outputs: `UI06-01..04`
- Result: all four commands stopped during TLS connection establishment with
`LibreSSL SSL_connect: SSL_ERROR_SYSCALL`.
- Provider task IDs: none returned.
- Output/report files: none created.
- Paid-generation claim: none; no submitted task or charge can be confirmed.
- Product/art judgment: none.
Owner subsequently authorized retrying transient image-generation failures.
Retry must use the same provider, prompts and references. Wait for a successful
no-cost reachability check before resubmission; if a later attempt exposes a
task ID, resume that task rather than creating a duplicate.
## Attempt 02 — `INVALID_INTERRUPTED_NO_TERMINAL`
- API reachability had recovered (`HTTP 404` from the service root).
- The `UI06-01` command lost its controlling execution turn before retaining
stdout, an output image or a report.
- Provider task ID: unknown; submission cannot be proved.
- Product/art judgment: none.
- Recovery: the installed image CLI was minimally repaired to persist and
print `task_id` immediately after a successful POST, before polling. Future
interruption can therefore resume a known task without blind resubmission.
@@ -0,0 +1,20 @@
# Phase 6 UI Direction Batch 01
This batch is visual-direction evidence only. It is not final art, a sprite
sheet, runtime UI, animation evidence or human acceptance.
| ID | Output | Task ID | Cost | SHA-256 | Status |
|---|---|---|---:|---|---|
| `UI06-01` | `outputs/01_battle_ui.png` | `task_01KZRVBA2KK7HHCZV3EYSPKY0K` | USD 0.014 / 0.14 credits | `bb002500e3f1897e6fc9a1b997bcbbd9322a2217667d3a132b555172ebb4e958` | `OWNER_REJECTED_STYLE_AND_SCALE` |
| `UI06-02` | `outputs/02_market_ui.png` | `task_01KZRVK5N103GBFGYX43TCZX70` | USD 0.014 / 0.14 credits | `4499a0030a79d653fe5c2d84ee99bf64d553194d7f765ba358b0ee4ce43756b3` | `OWNER_REJECTED_STYLE_AND_FIXED_IDENTITY` |
| `UI06-05` | `outputs/05_battle_ui_japanese_scale.png` | `task_01KZRW0K4H7K238ZYE6VF8D8RS` | USD 0.014 / 0.14 credits | `74c98f8270990168026633abebf046253892fca93068ddcd6cff9953977097db` | `OWNER_APPROVED_STYLE_AND_SCALE` |
Known accepted provider cost so far is USD 0.042 / 0.42 credits across three
completed tasks. Attempt 01 never reached submission. Attempt 02 produced no
retained task ID and remains unproven rather than counted as a product result.
`UI06-05` is the accepted calibration image. It targets ordinary unit height of
`10-14%` of the 2048x1152 canvas, a complete ten-member Company, materially
larger enemy groups, continuous horizontal entry space and a Japanese 2D
sprite-like presentation. It authorizes formal UI master exploration, not
engine development or bulk final asset production.
Binary file not shown.

After

Width:  |  Height:  |  Size: 3.9 MiB

@@ -0,0 +1,19 @@
{
"mode": "execute",
"model": "gpt-image-2",
"size": "16:9",
"resolution": "2k",
"output": "docs/prototype/generation/phase6_ui_direction_01/outputs/01_battle_ui.png",
"local_references": [
"docs/prototype/generation/01_ch01_battle.png",
"prototype/gallery/evidence/r8-phase5-attempt05/desktop-battle.png"
],
"remote_reference_count": 0,
"prompt_chars": 3368,
"task_id": "task_01KZRVBA2KK7HHCZV3EYSPKY0K",
"status": "completed",
"bytes": 4137015,
"sha256": "bb002500e3f1897e6fc9a1b997bcbbd9322a2217667d3a132b555172ebb4e958",
"cost": 0.014,
"credits_cost": 0.14
}
Binary file not shown.

After

Width:  |  Height:  |  Size: 3.6 MiB

@@ -0,0 +1,19 @@
{
"mode": "execute",
"model": "gpt-image-2",
"size": "16:9",
"resolution": "2k",
"output": "docs/prototype/generation/phase6_ui_direction_01/outputs/02_market_ui.png",
"local_references": [
"docs/prototype/generation/04_elements_board.png",
"prototype/gallery/evidence/r8-phase5-attempt05/desktop-market.png"
],
"remote_reference_count": 0,
"prompt_chars": 3295,
"task_id": "task_01KZRVK5N103GBFGYX43TCZX70",
"status": "completed",
"bytes": 3797621,
"sha256": "4499a0030a79d653fe5c2d84ee99bf64d553194d7f765ba358b0ee4ce43756b3",
"cost": 0.014,
"credits_cost": 0.14
}
@@ -0,0 +1,10 @@
{
"mode": "dry-run",
"model": "gpt-image-2",
"size": "16:9",
"resolution": "2k",
"output": "docs/prototype/generation/phase6_ui_direction_01/outputs/05_battle_ui_japanese_scale.png",
"local_references": [],
"remote_reference_count": 0,
"prompt_chars": 4603
}
Binary file not shown.

After

Width:  |  Height:  |  Size: 3.6 MiB

@@ -0,0 +1,16 @@
{
"mode": "execute",
"model": "gpt-image-2",
"size": "16:9",
"resolution": "2k",
"output": "docs/prototype/generation/phase6_ui_direction_01/outputs/05_battle_ui_japanese_scale.png",
"local_references": [],
"remote_reference_count": 0,
"prompt_chars": 4603,
"task_id": "task_01KZRW0K4H7K238ZYE6VF8D8RS",
"status": "completed",
"bytes": 3768490,
"sha256": "74c98f8270990168026633abebf046253892fca93068ddcd6cff9953977097db",
"cost": 0.014,
"credits_cost": 0.14
}
@@ -0,0 +1,13 @@
Create one original 16:9 landscape FORMAL GAME UI CONCEPT for Aetherbound Guild, a premium 2D Japanese hand-drawn fantasy auto-battler. It must look like an actual polished in-game battle screen, not key art, not a web dashboard, not a marketing mockup.
The supplied battle-art reference controls only the original hand-drawn line quality, bright mineral palette, weathered ceramic/brass/canvas materials, shallow side-view camera and windmill-valley atmosphere. The supplied prototype reference controls only the information topology: one horizontal Company line, urgent event at upper left, pause/speed/inspect at upper right, causal event rail near the lower edge. Do not copy exact geometry, names, text, or placeholder circles.
GAME WORLD: a bright unstable fantasy relay valley with windmills, field guild equipment and a clean shallow three-quarter side camera. The world occupies at least 70 percent of the screen. Six generated adult Recruits form ONE continuous horizontal line from protected rear-left to exposed front-right. They are replaceable run participants, not named protagonists: different bodies, faces, hair and repaired field clothing. Their fixed professions are readable through silhouette and tools: back-line channeler, healer, archer, polearm controller, shield guard, front-line duelist. Mature 5.5-to-6-head proportions, practical poses, no glamour lineup, no chibi.
ENEMIES: on the right, an original large ceramic wind-beast boss and three smaller civic-machinery creatures. The boss is winding up a coral danger projection toward the exposed front Recruit. The Company is visibly executing an automatic action chain: shield interception, healing thread and aether projectile. No manual attack buttons.
UI: original game chrome attached to screen edges and in-world guild tools. Chalk-white ceramic plates, charcoal ink strokes, repaired canvas tabs, weathered brass fasteners, small enamel color accents. Upper left: compact objective plate and one urgent-event strip. Upper right: icon buttons for pause, 1x/2x/4x speed and inspect, each with clear physical affordance. Each actor has a compact anchored health/condition marker using shape plus color. Bottom edge: a thin causal timeline with three event receipts and a separate unobtrusive retreat entry only inside a pause/tool zone. Show one coral target connector, one cyan protection connector and one gold committed-result marker. Controls are square or lightly rounded, never pill clouds. Keep safe margins and 48-point touch scale.
FOCAL ORDER: first the imminent coral attack and exposed front Recruit; second the complete ordered Company line and their automatic responses; third objective and playback/inspect tools. High readability at phone landscape size. Multi-hue semantic palette: paper white, charcoal, cyan, gold, coral, safe green, focus cobalt.
No readable prose, no fake Chinese, no gibberish typography, no logo, no watermark. Use empty label lines, short neutral bars, icons and numerals only; final text will be engine-rendered. No copied commercial layout, no dwarves, no fixed hero cast, no card wall, no three lanes, no square combat grid, no manual skill bar, no portrait-card roster, no dark dashboard, no photorealism, no 3D render, no western mobile gloss, no gacha rarity frames, no pixel art, no excessive particles, no bokeh, no gradient orbs.
@@ -0,0 +1,17 @@
Create one original 16:9 landscape FORMAL GAME UI CONCEPT for Aetherbound Guild, a premium 2D Japanese hand-drawn fantasy auto-battler. It must look like an actual polished integrated Market screen used repeatedly between battles, not a web shop, card wall, tavern menu or marketing mockup.
The supplied element-board reference controls only original hand-drawn line quality, mature practical people, repaired canvas, ceramic, enamel and weathered brass materials. Ignore its fixed named identities: this game uses generated replaceable Recruits. The supplied prototype reference controls only required information topology: Company bench, three Recruit offers, 12-slot run inventory, Recruit/Equipment/Risk/Sell modes, refresh, lock and one review action. Redesign the composition as game-world UI; do not copy its exact rectangles, names or text.
SETTING: a sunlit field-guild requisition hall open to an unstable fantasy route. A painted world remains visible through the workshop threshold. A long physical muster bench and requisition counter organize the screen. World and full/half-body figures occupy at least 62 percent of the usable surface.
COMPANY: a shallow upper-left bench shows two currently owned generated Recruits and two clearly open line-capacity positions. The owned figures are distinct adults with practical silhouettes and small profession tools. Capacity and Company are different visual facts.
OFFERS: three equally weighted generated Recruit offers stand at the counter as full or three-quarter figures, not portrait cards. Each has a profession emblem, one two-sided Trait seal, a compact stat-shape cluster, Coin price plate and one physical comparison tab. One offer is focused with a white-and-cobalt double contour, but none is labeled best. The player can inspect before committing.
INVENTORY AND MODES: a low 12-slot equipment tray spans the bottom, containing four readable physical objects and eight empty sockets. A compact vertical tool rail switches Recruit, Equipment, Risk and Sell modes using icon plus text-safe line. Refresh cost, next escalation, two lock pins, Coin and Replacement Credit appear as secondary edge tools. The only strong gold action is Review Focused Offer; purchase is not committed on first focus.
UI MATERIAL: chalk-white enamel, charcoal ink, repaired deep-green canvas, brass fasteners, small coral/cyan/gold/cobalt state accents. Square or 4-8px corner equivalents, restrained shadows, no floating glass. Use connected rails, trays and one attached inspect ledger rather than nested cards.
FOCAL ORDER: first the three materially different Recruit offers; second current Company and open capacity; third inventory, Coin, refresh/lock consequence and review action. Readable at phone landscape size with generous safe margins.
No readable prose, no fake Chinese, no gibberish typography, no logo, no watermark. Use empty label lines, short neutral bars, recognizable icons and numerals only; final text will be engine-rendered. No copied commercial layout, no dwarves, no fixed protagonist, no anime gacha portraits, no rarity glow, no portrait-card wall, no tavern cliche, no dark dashboard, no beige parchment monoculture, no rounded pill cloud, no photorealism, no 3D render, no pixel art, no bokeh or gradient orbs.
@@ -0,0 +1,17 @@
Create one original 16:9 landscape FORMAL GAMEPLAY UI CONCEPT for Aetherbound Guild, a premium 2D Japanese hand-drawn fantasy auto-battler. This image is specifically a correction of two rejected directions. It must look like a real side-view 2D game running at full-party scale, not a hero illustration, not key art, not a web dashboard.
ART STYLE — STRICT: clean Japanese 2D game illustration with thin confident ink lines, simplified four-to-five-head stylized adult proportions, flat cel-color shapes, restrained two-step shading, small selective highlights, and a softly painted gouache/watercolor environment behind crisp sprite-like actors. Bright readable fantasy colors, airy daylight and clear silhouettes. Characters are designed to animate as layered 2D cutout/sprite actors. The finish should feel like a polished Japanese console strategy RPG, not Western comic art. Avoid heavy black outlines, gritty leather realism, detailed pores, painterly oil rendering, bulky realistic anatomy, steampunk concept-art rendering, photorealism and 3D.
UNIT SCALE — STRICT: canvas 2048x1152. Ordinary allied and ordinary enemy figures must be only 120-165 pixels tall, approximately 10-14 percent of total screen height. A boss may be 260-320 pixels tall. Do not enlarge a hero for emphasis. Show a COMPLETE TEN-MEMBER generated Company at once. All ten allies remain individually readable through hair shape, color block, weapon/tool and profession silhouette, but faces do not need portrait detail at this scale.
BATTLEFIELD: one long continuous horizontal ground strip across a luminous aether meadow and ruined relay shrine. The camera is a clean side view with gentle depth layers, similar to a stage rather than a shallow cinematic hero lineup. Ten generated adult guild recruits travel from left toward right in one compact horizontal formation. Professions include shield guard, spear guard, duelist, archer, healer, mage, summoner, artificer, monk and beast warden. They are replaceable recruits, not named leads. On the right and center-right, show twelve ordinary magical enemies arriving in two readable clusters plus one larger wind spirit boss farther right. Leave visible entry space at both horizontal edges so more units could continuously enter. A few bodies overlap naturally in combat, but silhouettes and health markers remain distinguishable.
AUTOMATIC ACTION: no manual skill buttons. Show the line fighting automatically: one shield block, one small healing ribbon, one arrow volley and one enemy spell telegraph. Effects are compact and cannot hide units. The player reads preparation paying off rather than controlling attacks.
UI SCALE AND TOPOLOGY: world occupies about 72 percent of the frame. A slim top rail uses original enamel-and-cloth fantasy UI, with allied count/progress at left and enemy count/progress at right, plus small pause, speed and inspect icon tools between or at the edge. Bottom area uses a low 18-20 percent roster/status rail with ten compact member markers arranged in two tidy rows of five or one dense row where readable. Each marker shows a tiny profession portrait/silhouette, health/mana/condition bars and one status shape. Keep the battlefield visible behind or above the rail. Timer at lower left; retreat only as a small separated flag tool; no giant cards. Square or lightly rounded controls, 48-point touch targets, no pill cloud.
COLOR AND MATERIAL: lively Japanese fantasy palette rather than brown steampunk — sky blue, spring green, ivory, charcoal, vermilion danger, turquoise magic, leaf green recovery, warm gold commitment and cobalt focus. UI may use cloth tabs, pale lacquer, simple brass pins and dark ink, but brass/leather must not dominate.
FOCAL ORDER: first the dense clash between the ten-member Company and incoming enemy group; second the readable small-unit professions and current dangerous telegraph; third the compact allied/enemy counts and roster health state. The overall read must be a scalable army-building auto-battle, not six large heroes posing.
No readable prose, no fake Chinese, no gibberish text, no logos, no watermark. Use icons, bars, numerals and empty label space only. No copied commercial layout or assets, no dwarves, no fixed protagonist, no relationship framing, no card wall, no multiple lanes, no grid, no manual attack or skill bar, no gacha portraits, no huge allied figures, no dark cave, no Western comic rendering, no gritty steampunk, no photorealism, no 3D, no pixel art, no chibi baby proportions, no excessive bloom, particles, bokeh, gradient orbs or cinematic depth blur.