Lab · The Atlas

The recipes behind the work.

Each capability ships with a tier, a cost, a sample, a verdict — same discipline as the work. Methodology is the moat, not the tools.

11 axes · 71 capabilities · 47 with project sample · 24 forthcoming / external · atlas v1
§1

Storyboard Formats

image-layout-end · pre-viz / collage / sheet
14 items
3×3 Narrative
3×3 Narrative
1:1
Plot-driven 3-act storyboard · 9 panels · single image · pre-viz for Comic Skit / Narrative MV
stray-cat watercolor · 9/9 character lock · 6 distinct locales
validated_2plus ¥0.5
prompt id: 3x3_narrative_v1_stray_cat_anime view detail →
12-Panel Rhythm
12-Panel Rhythm
4:3
Mood-driven multi-angle collage · K-pop / fashion / brand reel · feed as i2v ref (¥0.5 collage → ¥3 video)
13-Layer commercial-vertical skeleton · yuan.studio brand reel hero source
observed_once ¥0.5
view detail →
16-Panel Day-Cycle
16-Panel Day-Cycle
4:3
Long narrative flat-list · multi-environment + lighting evolution (morning → night) · slice-of-life
Eclipse Day-in-Life · 6 env clusters · 3 wardrobe states · 4 lighting stages
observed_once ¥0.5
view detail →
2×5 Cinematic Shot List
2×5 Cinematic Shot List
2:1
长 prompt + per-cell shot spec + 风格 anchor (BBC Planet Earth / 影像类) → 10-cell 电影分镜表 · 每 cell 是真电影 shot (景别 / 灯光 / subject behavior 显式 spec) · 接 i2v 出 15s 短片
项目内首验证 2026-05-01 · 38s wall · 4 hardest tests + 2 bonus 全 PASSED. **6th axis-of-variation** in storyboard-grid-formats pack: cinematic shot-list (vs narrative beat / transformation / style / rhythm / time). 首样本: seed→tree 15s short film · BBC Planet Earth 美学 · 用户引用 (Stage 1 of 2-stage pipeline · Stage 2 = Kling 3.0 · 见 external_intel/kling_3_4k_15s_2026_05.md). **4 hardest tests 验:** (1) Scale shift 同图内: aerial → macro → underground → wide → mirror · 全 coherent (2) Subject continuity across 10 morphs: 同主体 (seed/sprout/tree) 不同生命阶段 (3) Mirror composition: frame 1 + 10 同机位回环 (4) Time-lapse → 1 frame: cell 7 自拆 vertical strips 显季节渐变 (创意解 · GENIUS) **2 bonus 发现:** (a) 模型自加 cell 编号 1-10 · 0 prompt 提示 · production 工业级 storyboard 工具感 (b) Storm cell (5) 自切 cold blue-grey vs 其他 warm · 模型自推 phased lighting **Pipeline pairing:** Stage 1 (此 capability) + Stage 2 (Seedance 3 段 5s + ffmpeg + Topaz 4K) = 复刻 Kling 4K-15s 路径. 参 pipeline-collage-i2v-rhythm (12-panel · 5s 输出 · cell→beat ~0.4s) vs 此条 (10-cell · 15s 输出 · cell→shot 1.5s). 同 pattern · 不同 cell-time 比. Sister storyboard-3x3-narrative (同 narrative axis · 但 9 cells / story beat 而 此条 10 cells / film shot). Sister storyboard-12panel-rhythm (同 collage-i2v ref pattern · 但节奏点 vs 此条电影 shot).
observed_once ¥0.5
view detail →
4×4 Transformation
4×4 Transformation
4:5
Single-env + progressive accumulation + typography-as-feature · save-distribution (Pinterest/IG/RED) · GRWM / makeup / cooking
GRWM Outfit · luxury editorial · 4-cluster dressing arc · 4:5 ApiMart 首验
observed_once ¥0.5
prompt id: 4x4_transformation_v1_grwm_outfit view detail →
5×5 Style Grid
5×5 Style Grid
1:1
单 subject + 25 cells × distinct styles · style-axis (vs narrative/rhythm-axis) · style 探索 / 客户教育 / pre-viz · minimal prompt unlock model prior
项目内首验证 2026-05-01 · 最大 grid (25 cells) + 最短 prompt (10 词). 新 axis: style-axis (vs narrative-axis 3x3 · rhythm-axis 12-panel · time-axis 16-panel daycycle). Subject 锁定 · style 变化 · 同 fox 25 种 anime/cartoon/illustration 风格. 5 unique 维度 (全 PASSED 4/5 · 1 NO): (1) 5x5 layout · 25 cells · model 上限 ≥25 · cells 比例均匀 (2) Minimal prompt → rich output · prompt-is-coordinate 经典 demo (3) ~18-20 RECOGNIZABLY distinct styles (5 partial overlap on cyberpunk/cute mascot) (4) Subject auto-consistency (model 选 fox + 锁 waving pose) (5) Auto-labels NO · 无文字标签 (要 label 须显式 'with style names below') **GOTCHA: 14 min wall** (855s actual vs 100s estimate · 8.5x) · 25-cell grid 是 ApiMart/OpenAI 资源密集 · queueing 严重 · cost still ¥0.5 但 wall 30-50x 普通 generation · ApiMart polling 8s + 360s deadline 对此 task 不够 · production 跑此 cap 须 timeout ≥20min Production use: · style 探索工具 · ¥0.5 看 model 25 种风格 prior on a subject · 客户教育 ('我们风格库' baseline) · 风格对比研究 · pre-viz 风格选项 Sister storyboard-3x3-narrative (axis: narrative · 单 subject 不变 · 9 个 narrative beat) + storyboard-12panel-rhythm (axis: rhythm · brand reel · 12 panel asymmetric).
observed_once ¥0.5
view detail →
All-in-One Asset Sheet
varies
多元素 1 张图 (主角 + 配角 + 武器 + 战马 + 环境) · 替多次生图 · 风格统一
适合古装 / 战争 / 史诗 · 1 外部样本(项羽乌江战役)
observed_external · external only ¥0.5
view detail →
Storyboard Treatment Ref
varies
70% 视觉 + 30% sidebar specs (时间戳 / 相机 / SFX / BGM) · negative-style 引用作 R2V structural ref
适合多分镜混剪 / 广告 / 古风短片
observed_external · external only ¥0.5
view detail →
Staging Blueprint Ref
varies
俯视平面图 + 颜色站位 + 运动轨迹 · 4 ref 分工 · multi-character 站位锚
~70-85% 跟随度 · 不建议新手
observed_external · external only ¥0.5
view detail →
360° Equirectangular Panorama
360° Equirectangular Panorama
2:1
4096-class 360 全景 · seamless horizontal wrap · undocumented ApiMart 2:1 aspect
1774×887 · 名字艺术家滤拦 · L/R RGB diff 1.09 (无缝)
observed_once ¥0.5
view detail →
Motion Sheet Recreation
Motion Sheet Recreation
1:1
Monochrome 教学 sheet 重画成 color photoreal · 同 layout/numbering/箭头 + character/wardrobe/color 三重替换
16-panel layout 100% 复刻 + character 100% 替换
observed_once ¥0.5
view detail →
Production Bible · 12-Panel + Cast + Props + Location
Production Bible · 12-Panel + Cast + Props + Location
16:9
Full pre-production bible · 12-panel storyboard + cast turnaround + prop sheet + location refs + style refs + palette in one GPT Image 2 call · pre-viz / pitch deliverable
Promoted via tools/promote · 2026-05-02. Source probe sidecar: experiments/production_bible_probe_v1_20260502/production_bible_probe_v1.run.yaml
observed_once ¥0.5
view detail →
Ring Spectrum Poster · Center + Ring · 9:16 Mobile-First
Ring Spectrum Poster · Center + Ring · 9:16 Mobile-First
9:16
Center anchor + ring of N same-character variants in continuous immersive space · 9:16 mobile-first viral poster · system prompt + hidden reasoning + audience CTA paradigm (per 岳哥 4-version system)
Promoted via tools/promote · 2026-05-02. Source probe sidecar: experiments/maltese_performer_ring_spectrum_v1_20260502/maltese_performer_ring_spectrum_v1.run.yaml
observed_once ¥0.5
view detail →
3×3 Expression Sheet · Torn Paper Circles · Pixar 3D
3×3 Expression Sheet · Torn Paper Circles · Pixar 3D
1:1
9-emotion × 9-outfit cross-axis character bible · torn paper circle frames(non-rectangular panel · 新 visual primitive)· Pixar 3D style · enumerative descriptive sub-paradigm · single GPT Image 2 call
Promoted via tools/promote · 2026-05-02. Source probe sidecar: experiments/pixar_expression_sheet_verbatim_v1_20260502/pixar_expression_sheet_verbatim_v1.run.yaml
validated_2plus ¥0.5
view detail →
§2

Multi-Ref Modes

image_urls 编辑模式 · ref-driven generation
6 items
Basic T2I
Basic T2I
any
Pure text-to-image · 无 ref · 最简单的入口形态
kitten watercolor + srai_poster + cav_poster 等多次
production_repeatable ¥0.5
view detail →
2-Ref Scene + Character Composition
2-Ref Scene + Character Composition
1:1
1 场景 ref + 1 角色 ref → 角色置入场景 · 场景 1:1 复刻 + 角色 identity 锁
1996 春节 + yuan_kid 真儿童 ref · OpenAI minor filter 通过
observed_once ¥0.5
view detail →
5-Frame Charsheet Generation
5-Frame Charsheet Generation
4:3
单 ref → 多角度 / 多表情 / 多服装 charsheet · turnaround 替代品
kitsunebi 验证 · 比 turnaround grid 更适合作 R2V multi-ref input · 信息密度 5 panel (vs 7-pose DSLR sister 7 panel)
validated_2plus ¥0.5
view detail →
7-Pose DSLR Character Sheet
7-Pose DSLR Character Sheet
4:3
Pure T2I · 7-panel 工业级 character bible (3 face profiles top · 4 body profiles bottom) · DSLR aesthetic · 直接进 IP refs/ · 适合 R2V multi-ref + character lock
项目内首验证 2026-05-01 · insmind_com 2-stage pipeline image part. Layout 严格命中 · 7 panel = 3 face (front/45°/side) 上行 + 4 body (front/45°/side/back) 下行. Outfit lock 完美跨 7 panel (olive bomber + white tee + 深 jeans + 白鞋). Identity 7/7 同人 (Asian male default · 因 prompt 没指定性别). DSLR aesthetic 兑现 (软光 · gradient 白底 · photoreal). Fault-tolerant: 'white tea' typo 仍渲染 white tee. KEY INSIGHT: pure T2I + 文字 anchor 已能产出工业级 charsheet · 不需要 ref input. LIMITATION: 没指定性别 → default 男 · multi-stage pipeline 必须 stage 1 提前 lock 角色 anchor (i.e., 加 'female named Sarah' 当 stage 2 video 角色是女). Sister · multiref-5frame-charsheet (5 panel · validated_2plus on kitsunebi anime aesthetic).
observed_once ¥0.5
view detail →
Multi-Ref Real-Minor IP
Multi-Ref Real-Minor IP
varies
真儿童照 ref + multi-ref edit mode · ApiMart relay 通过 OpenAI minor filter
5+ sample 实测通过 (1996 chunjie scene · business headshot v1/v2/v3 · NYC vlogger v2) · 跨 context (历史 / 商务 / 旅行 vlogger) · descriptor cluster ('attractive/beautiful/influencer') 加 minor 仍 PASS · 推 production_repeatable 候选
validated_2plus ¥0.5
view detail →
Single-Image Landing-Page Composite
varies
1 张图含 nav / hero / cards / footer 4 区 · commercial UI ad pre-viz
0 项目内 · 1 外部 (DStudioproject Higgsfield)
observed_external · external only ¥0.5
view detail →
§3

Signature Aesthetics

director's-taste assertions (Track 2) · 不是 empirical claim
4 items
SRAI · Semi-Realistic Animation Illustration
SRAI · Semi-Realistic Animation Illustration
1:1
Arcane × Klaus × Wolfwalkers fusion · 暗调室内戏剧 · ink linework over gradient washes
cat_dad 绑定 · §16.1 · 用功能性词汇代替 named-artist
observed_once ¥0.5
prompt id: 3x3_narrative_v2_cat_dad_hr_srai view detail →
CAV · Cinematic Anime Vista
CAV · Cinematic Anime Vista
1:1
Makoto Shinkai × Ghibli epic 户外 · orange-teal grading · golden hour · wildflowers
yuan_kid 绑定 · §16.2 · 首次 submit gateway timeout (10min) 可能发生
observed_once ¥0.5
view detail →
PIX3D · Pixar-Adjacent 3D CGI
PIX3D · Pixar-Adjacent 3D CGI
1:1
合家欢 3D CGI tone · 5 anchor 不写 'Pixar' 避 copyright filter · 双角色 identity + 3D fidelity 双 lock
Seedance 出片 fidelity 反超 SRAI · 训练 bias 对齐 Pixar 多于 painterly
observed_once ¥0.5
view detail →
SNAPSHOT_CANDID · 失败美学伪手机日记
SNAPSHOT_CANDID · 失败美学伪手机日记
4:5
不是 cinematic · "失败美学 · 伪手机日记" · 同步解锁 12-frame collage 格式
§16.4 · 3-sample 演化 · hypothesis 已升级 (2026-05-01). Sample 1 · v1 cafe rainy window: partial · 中性 anchor ('phone diary / handheld tilt') 输给 polish prior. Sample 2 · NYC vlogger: partial · 中性 anchor ('low-quality smartphone') 输给 polish prior. Sample 3 · pathetic redraw: **BREAK-THROUGH** · EXTREME anchor ('pathetic / clumsy / MS Paint / pixel-by-pixel' + self-mocking closer) 击穿 polish prior. **新 hypothesis · polish ceiling 可破 · 但需要 INSULTING / EXTREME DEGRADATION anchor · 中性描述不够**. Implication for SNAPSHOT_CANDID v2 · 改写 anchor 用 extreme 形容词 ('utterly amateur / pathetic phone snap / shamefully unpolished') 应能突破 · 待跑.
observed_once ¥0.5
view detail →
§4

Aesthetic Anchors

commercial-vertical 美学锚 (跨 IP · 可借) · §19 派生
3 items
KIDPOP_BRIGHT
KIDPOP_BRIGHT
4:3
儿童 idol vertical · §19.2 派生于 EDITORIAL_GLOSS · 4 sample 全 PASS Gate 1
image-end production_repeatable · video-end observed_once
production_repeatable ¥0.5
view detail →
EDITORIAL_GLOSS
varies
成年 K-pop / fashion / luxury 商业 vertical · borrowable
§19.1 · 0 项目内 · 1 外部 (K-pop · Rita 平台)
observed_external · external only ¥0.5
view detail →
SCI_FI_MILITARIZED_GLAM
SCI_FI_MILITARIZED_GLAM
1:1
AI Punk Studio cyberpunk commercial · Aitana Lopez 类 AI influencer · 5 anchor + body emphasis
§19.3 · 4 项目内 sample · character/wardrobe/aesthetic 一致性强
validated_2plus ¥0.5
view detail →
§5

Composition Primitives

prompt 物理写法 · 跨 capability 通用
4 items
Bracket-Prefix Block Header
[STYLE] / [LAYOUT] / [NEGATIVE] 等 · 替 inline `Style:` 散文式
0 项目内 ablation · 1 外部 (Pollo AI 武术 motion-sheet)
observed_external · external only
view detail →
NEGATIVE Sub-Block
多 negative item 拆行 · 替 inline `Negative prompt:` 一行
PROMPT_ARCHITECTURE §2.11b · 待项目内 ablation
observed_external · external only
view detail →
Per-Cell Dual-Layer Caption
top-left title + bottom-left 3-4 line description · 替单层 caption · 适合 motion sheet
对 Seedance i2v motion alignment 影响待测
observed_external · external only
view detail →
Trigger Phrase · Convert each panel from N to M
Convert each panel from N to M of @imageN into a [content] beat · sheet → i2v 唯一已知触发
feedback_motion_sheet_as_i2v_primitive.md · 描述性短语 model 会 ignore
validated_2plus · forthcoming
view detail →
§6

Pipeline Integrations

gpt-image-2 作上游 · 接 Seedance 等 video provider
3 items
3×3 → Seedance Pre-Viz Pipeline
varies
¥0.5 storyboard 探 premise → ¥3-5 5s probe → ¥13-15 full · 平均省 35%
feedback_3x3_to_seedance_pipeline.md · cat_dad_vase_skit 端到端 validated
validated_2plus · forthcoming ¥0.5 + ¥14
view detail →
Collage-as-i2v with Rhythm Intent
varies
12-panel collage 喂 Seedance i2v · 'Convert each frame...connect them' · 5s ~6 beats
3 项目内 + 1 外部 · 跨 mood/IP/N 验证完整
production_repeatable · forthcoming ¥0.5 + ¥3
view detail →
Sheet-as-i2v Primitive
1:1
16-panel motion sheet 喂 i2v · 5s 抽样 ~50% panels · sport / dance / yoga sequence
5s ~50% panel 抽样是 i2v fundamental ceiling
validated_2plus · forthcoming ¥0.5 + ¥3
view detail →
§7

Banana Recipes

banana (gemini image) 专属 recipes · style sheets / edit fallback
4 items
Banana Character Sheet
varies
Single-character turnaround / multi-pose sheet · banana (gemini image) edit mode
Production-tested across multiple IPs · canonical sample to be selected
validated_2plus · forthcoming ¥0.5
view detail →
Banana Expression Sheet
varies
Multi-expression grid (joy/anger/sad/etc.) · banana edit mode · pairs with character_sheet
Pairs with character_sheet for full IP turnaround
observed_once · forthcoming ¥0.5
view detail →
Banana Aerial Xianxia
16:9
Aerial 仙侠 vista style recipe · banana-tuned prompt for 修真 vertical
仙侠 / 修真 vertical-specific style
observed_once · forthcoming ¥0.5
view detail →
Banana Universal Edit Fallback
any
When MJ output needs targeted edits (face/wardrobe/pose) · banana_gemini edit is universal fallback
feedback_character_family_resemblance_via_banana_ref · 1 ref → unify wardrobe across batch
validated_2plus · forthcoming ¥0.5
view detail →
§8

MJ Recipes

Midjourney v7 专属 recipes · oref 规则 / photorealism / 风格
4 items
MJ V7 Photorealism Stack
varies
IMG_ hack + raw + s50 + film anchors + lighting check · MJ v7 photoreal default
Multiple IPs validated · see recipes/image_gen/mj/styles/_index.md for the stack
production_repeatable · forthcoming varies
view detail →
MJ --oref Scale Rule
varies
close-up ow=400-600 锁脸 · wide 任 ow 只锁 wardrobe · 2026-04-19 MJ × Ghibli 验证
很可能适用 Seedance R2V 多 ref · cross-provider hypothesis
validated_2plus · forthcoming varies
view detail →
MJ --oref Single Real Photo Wins
varies
不用 turnaround grid · 不用 banana 编辑版 · 单张原始 close-up 真人照效果最佳
2026-04-19 Ocean Gaze 对比验证 · counter-intuitive (more refs ≠ better)
validated_2plus · forthcoming varies
view detail →
MJ Wide+Close Composite oref (Aitana trick)
varies
Extreme face close-up + wide body 拼成单图 → 上传作 --oref · 强迫 MJ 优先 macro face 细节
AI influencer 社区标准做法 · 2026-04-20 Ocean Gaze Shot 3 验证
observed_once · forthcoming varies
view detail →
§9

Seedance Recipes

Seedance 2.0 专属 prompt techniques · 3件套 / quantified motion / multi-shot / first-last
4 items
Seedance 3-Piece Prompt Format
[Camera+mm+motion] [Action+env] [SFX] / shot · 比 5-layer 紧 · 适合 multi-shot
显式 mm 值精度比 'close-up/wide' 高 · 配 8-shot template 用
observed_once · forthcoming
view detail →
Seedance Quantified Motion Specs
"15cm shoulder punch" / "25° hip drop" / "180ms whip" 量化 beat 抽象 adjective · 适用 dance / 物理交互
Timeline + quantifier = 手写 Gemini Mode A 等价
validated_2plus · forthcoming
view detail →
Seedance Multi-Shot Packing
`Shot 1/2/3:` 标记被尊重为硬剪辑点 · 5s 可塞 3 shots × 1.7s · 视觉密度翻倍
字幕/音频会被归一化成纯 CN · mixed-register EN loan words 丢失 · trade-off 见 memory
validated_2plus · forthcoming
view detail →
Seedance First/Last Frame Interpolation (Path 4)
any
Ark direct content[] schema · role first_frame + last_frame · 起止帧像素直接复制 · 中间 ~3s morph
无 face averaging · 无 stylization decay · ApiMart 不支持 role · 必须 Ark direct via scripts/seedance_first_last.py
validated_2plus · forthcoming ¥3
view detail →
§10

Templates

参数化 prompt 产品 · 用户填占位符 / 加 ref · 商业化路径明确 · 商务头像 / NYC 旅行 / GRWM 等
24 items
NYC Travel Vlogger · No-Ref (Mood Vibe)
NYC Travel Vlogger · No-Ref (Mood Vibe)
1:1
9-panel mood-driven travel collage · pure T2I · 模型造虚构 vlogger · creative pre-viz / mood vibe / 灰区"假旅行账号" 用途 · JSON-structured prompt · landmark precision · polished influencer aesthetic (NOT phone-diary)
项目内首验证 2026-05-01 · NYC sample · 9/9 landmarks legit (Statue of Liberty/Times Square/Central Park/Joe's Pizza/Matisse Dance/AMNH T-Rex/Brooklyn Bridge/Empire State/SoHo). Identity consistency 跨 9 panel (虚构 model · 模型自动锁) · JSON-structured prompt parsed 精准. BUT prompt 自我宣称 'unpolished phone diary' 没 deliver (实际 polished Instagram feed). Cross-references signature-snapshot-candid '模型 polish ceiling' 假说 (现 2-sample 支撑). Sister · templates-nyc-travel-vlogger-ref (有 yuan_kid ref 版 · 1 张真人 → 9-panel 'you in NYC'). 原 id storyboard-3x3-rhythm-travel-template (Option C refactor 2026-05-01 · 移到 templates 轴).
observed_once ¥0.5
view detail →
NYC Travel Vlogger · Personal (with Ref)
NYC Travel Vlogger · Personal (with Ref)
1:1
1 张真人照 + JSON multi-scene prompt → 9-panel "你在 NYC 旅行" 海报 · identity 跨 9 panel 保 + landmarks 9/9 准 + NO age-up · personal branding / LinkedIn / 母亲节礼物 / 童装 marketing
项目内首验证 2026-05-01 · 4yo yuan_kid + 同 JSON prompt + 'Identity Do not age up' clause. 4 axis 全 PASS (identity 跨 9 panel 完美 · NO age-up · landmarks 9/9 · minor filter 通过). Ref 主导 prompt-described outfit (粉卫衣 + Hello Kitty 替代 'fashionable travel outfits') · production-good. SNAPSHOT_CANDID 仍 partial (polish ceiling 持续 · ref 不影响). 解锁 commercial use case · ¥0.5/张 personal travel feed · 任何用户照可入. Sister · templates-nyc-travel-vlogger-noref (无 ref T2I version). 原 id identity-transform-3x3-personal-travel-vlogger (Option C refactor 2026-05-01 · 移到 templates 轴).
observed_once ¥0.5
view detail →
Porcelain Sculpture Portrait Template
Porcelain Sculpture Portrait Template
1:1
真照 + selective identity copy prompt → 3D relief mixed-media 雕塑 (porcelain + rose-gold + pearl filigree) · 显式 disentangle structure-vs-pose · gallery / gift / luxury brand collateral
项目内首验证 2026-05-01 · yuan_kid 4yo · 9 anchor 0 漏 PASS. 重大 framework 新发现 · **selective identity copy 机制**: 之前 templates 假设 copy whole identity · 这条显式 disentangle: ✅ Copy: structure + proportions + identity markers (五官/比例/骨架) ❌ Don't copy: expression + pose + angle Model 真做到 (源 front-facing + 张嘴 → 输出 3/4 侧 + 闭眼 serene · 五官准). Implication · gpt-image-2 内部有 identity-disentanglement 能力 · 之前没用. Production rule · 想精确控制 ref 哪部分 copy / 哪部分换 · 用显式 'use ref for X only' 子句. Anchor conflict resolution · "of a woman" vs 4yo ref: Model 升 age 到 young woman (~16-20) · 保 face 结构 · 因为 prompt 没禁 age-up. 对比 business-headshot ('do not change identity' explicit) → 没 age-up · 仍 4yo. Production rule · 想锁年龄需显式 'preserve age / do not change' · 否则 model 自由 interpolate. Sister · pencil-sketch + pet-crayon (medium transform 同族) · 但此条更复杂 (mixed-media 3D relief vs 单 medium 2D).
observed_once ¥0.5
view detail →
Surreal Mural Portrait Template
Surreal Mural Portrait Template
1:1
复合 prompt (subject + 超现实 merge + 环境 mural + cursive 文字) → suburban 墙上的写实壁画 · subject 融入 environment · 街景 + 玫瑰 + 泡泡糖 + Instagram-ready
项目内首验证 2026-05-01 · pure T2I · 6 anchor 完美 PASS + 智能解 anchor conflict. NEW prompt 范式 · environmental composition (subject IN scene) · 不同于之前 7 templates 都是 portrait composition (subject 占主 frame). 关键 framework finding (第 2 个 anchor conflict 智能解): '8K' anchor 不是 '8K real photo' 强约束 · 是 'highest detail in chosen medium' 相对约束. 被 medium anchor 拉去显示 'medium 内 8K 水平'. Sample · '8K + photorealistic mural painting' → 8K-detail mural (不是 8K photo). Sister 例 · '8K + visible pencil strokes' → 8K pencil work (不是 8K photo). Surreal merge (头发→玫瑰 cascading) 完美 honor · model 强 surreal compositional prior. In-image text 'Noor 🌸' cursive 粉 + emoji 都准 (短 string · ≤5 chars 安全区间). Production use: 个人品牌 / 街头艺术 marketing / 自定义壁画 mockup / Instagram 引流 · 任何 name 可换.
observed_once ¥0.5
view detail →
Pet Crayon Poster Template
Pet Crayon Poster Template
1:1
宠物名 + 复合 prompt → 蜡笔/naive sketch 风海报 · centered + warm yellow + vintage print + 中文 title · 礼物/SNS 头像/宠物 marketing
项目内首验证 2026-05-01 · 菜菜 (Siamese/Ragdoll cat) pure T2I · 6 anchor PASS + 1 limitation 暴露. Medium transform (crayon) 跟 pencil sketch 同样工作好 (gpt-image-2 medium prior 强). Editorial poster layout (centered + empty space + vintage print) 全 honor. Cat species accuracy 高 (point-coat coloring + blue eyes 准). Limitation · 长 CN 段落 (>15 字) hallucinate · 笔画像但不是真字 ('small descriptive text below' 翻车). 短 CN title (2 字) 完美. Production rule · CN text-in-image 用 short label / title · 不用 paragraph. Sister · templates-pencil-sketch-portrait (同 medium-transform 范式 · subject 人 · layout 肖像 vs 此条 subject 动物 · layout 海报含文字). 待跑变体 · (a) 用真 cat 照 ref (用户 workflow) · (b) 不同 medium (watercolor/油画 pet poster).
observed_once ¥0.5
view detail →
Photo Annotation Overlay · JP Vlog Style
Photo Annotation Overlay · JP Vlog Style
1:1
真照 + 中文 annotation rules → 源照 100% 保留 + 手绘 doodle 注解 + 日记 tone 短句 + Japanese-kawaii aesthetic · 小红书/RED viral vlog 模板
项目内首验证 2026-05-01 · yuan_kid 户外吃零食 · 9 段中文注解全 honor. 这是 6 种 prompt 范式中**最少破坏性**那个 (源照 0% 重绘 · 只 overlay). Model 超越 prompt · 主动加 emotional layer ('妈妈在旁边 默默照顾我 好安心'). 唯一瑕疵 · 把 fried snack 标作 '披萨' (annotation rule 对 · object 识别错). Production-grade 小红书 post · 0 post-editing · 直接发. 高商业价值 · ¥0.5/张 · viral SaaS 模板 (上传你的照 + 一键 = vlog post · 9.9 元). 6 种 prompt 范式光谱: annotation overlay (this · 0% destruction) ↓ medium transform (pencil sketch · ~30%) ↓ personalization (NYC ref · ~50%) ↓ identity transform (business headshot · ~70%) ↓ anti-polish redraw (pathetic · ~90%)
observed_once ¥0.5
view detail →
Oriental Ink Wash Painting Template
Oriental Ink Wash Painting Template
1:1
东方水墨 template (含 [SUBJECT][COLOR1][COLOR2] 占位符) → 真 水墨画 · 自动补印章 + 落款诗 + 远山 + 松 · museum-quality · 工艺品 / 装饰画 / 文化 marketing
项目内首验证 2026-05-01 · 填 [SUBJECT]=swordsman · [COLOR1]=ink black · [COLOR2]=mist grey · 8 anchor PASS + 3 bonus 自补. 重大发现 · Model 强 "Chinese 水墨 painting" **整套文化 convention prior**: Prompt 没要 · model 自补 印章 + 题画诗 (4字×2 "仗剑天涯 笑看风云") + 远山 + 松树 → 不只是 medium prior · 是 **cultural genre convention prior**. 类比 · 给 'Renaissance oil portrait' anchor 应也会自补 dark background + 服装层次 + 古典签名等 (待验证). Medium-transform 子族 **现 4-sample 全 PASS** (pencil 西graphite + crayon · watercolor 西 + ink-wash 东) · hypothesis "gpt-image-2 medium prior 强" 推 validated_2plus 强证据. 双用途 · 此 sample 同时: (1) 测 template prompt 范式 (2) 给 IP ink_wash_swordsman 补 canonical_portrait (之前是 placeholder)
observed_once ¥0.5
view detail →
Watercolor Landscape Template
Watercolor Landscape Template
1:1
风景 prompt → loose expressive watercolor 风景画 · 多元素平衡 (建筑+自然+海+天空) · gallery-worthy 油画质感 · 印刷品/家居装饰/旅游 marketing
项目内首验证 2026-05-01 · pure T2I · landscape · 9 元素 + 3 bonus 全 honor. Model 自补 sailboats + 海鸥 + 艺术家 signature · 增加 scene 真实感. 'Loose expressive watercolor' 真 honor (vs 'highly detailed' anchor 时 polished). Medium-transform sub-family **现 3-sample validated** (pencil 人 · crayon 动物 · watercolor 风景): 跨 3 subject 类 (人/动物/风景) + 3 medium (graphite/crayon/watercolor) 全 PASS · "gpt-image-2 medium prior 强" hypothesis 由 1-sample 升 validated_2plus tier. Production use: 印刷品 / 家居装饰 / 旅游 affiliate / 城市旅游 marketing · 替换风景描述即换地点.
observed_once ¥0.5
view detail →
Pencil Sketch Portrait Template
Pencil Sketch Portrait Template
1:1
JSON-structured prompt → 真 graphite-on-paper portrait · 7 anchor 全 honor (subject/pose/hair/clothing/medium/strokes/paper grain) · production-grade illustration · gift / commission / SNS / brand
项目内首验证 2026-05-01 · pure T2I · JSON-structured prompt + 7 anchor. 关键发现 · '8k highly detailed' anchor 与 'visible pencil strokes' 不冲突 · 模型解读为 'highly detailed PENCIL work' · 不是推回 polish. Framework insight · medium transform (pencil/watercolor/oil) ≠ anti-polish (diary/amateur). 前者 model prior 强 · 好做 · 后者 model prior 弱 · 需 EXTREME anchor 才破 (见 templates-pathetic-redraw vs signature-snapshot-candid 对比). 待跑变体 · (a) 加 yuan_kid ref 测 personalization 路径 · (b) 换 medium (watercolor/charcoal/oil) 验证 medium-transform 跨 medium 同样强.
observed_once ¥0.5
view detail →
One-Line Gesture Redraw Template
One-Line Gesture Redraw Template
1:1
单 ref portrait → 单条连续黑线 (不抬笔) · loose / 负空间大 / 极简姿态画 · pack avatar-redraw-2026q2 variant
项目内首验证 2026-05-01 · 48s wall · pack avatar-redraw-2026q2 variant 2/7. 单条连续黑线 · 跨整图 drift · 主体仅 loose 暗示 (脸 / 手 / 座椅 / 窗) · 大量负空间. 模型 honor "do not lift the pen" 约束 · 出片真有 1-line gesture 感.
observed_once ¥0.5
view detail →
MS Paint Stick Figure Redraw Template
MS Paint Stick Figure Redraw Template
1:1
单 ref portrait → 童画 stick-figure / MS Paint mouse-drawn 感 · 跟 pathetic 同祖 但显式 stick limbs · pack avatar-redraw-2026q2 variant
项目内首验证 2026-05-01 · 39s wall · pack avatar-redraw-2026q2 variant 3/7. 跟 pathetic-redraw 同 polish-ceiling 破解模式 (童画 / 失败美学) · 但显式 stick limbs · 出片含粉/绿/蓝/黄 bright color (vs pathetic 单色).
observed_once ¥0.5
view detail →
Torn Paper Collage Redraw Template
Torn Paper Collage Redraw Template
1:1
单 ref portrait → 撕边彩纸拼贴 · 不齐叠层 · 跨 medium 物理材质 · 出片 gallery-quality · pack avatar-redraw-2026q2 variant
项目内首验证 2026-05-01 · 57s wall · pack avatar-redraw-2026q2 variant 4/7. **本 pack 中 fidelity 最高之一** · 模型 honor 撕边纸 / 叠层 / 不齐 / 多色调 · 出片 gallery-tier 拼贴艺术. Identity 保留度 vs pathetic/one-line 高很多 · 这是 medium-transform 而非 polish-ceiling 破. Production 价值: 客户 portrait collage 替代 · 替 ¥几k 实体拼贴.
observed_once ¥0.5
view detail →
Notebook Margin Doodle Redraw Template
Notebook Margin Doodle Redraw Template
1:1
单 ref portrait → 圆珠笔走神课堂涂鸦 + 自动生成 annotation/arrows/side jokes · 模型 invent text · pack avatar-redraw-2026q2 variant
项目内首验证 2026-05-01 · 57s wall · pack avatar-redraw-2026q2 variant 5/7 · **pack 中 best**. **3 个新发现:** (a) **模型自动 invent text annotation** 极强 · 我们 0 prompt 提示具体内容 · 出片含 10+ 自创 quip: "hair = chaos" / "fuel" / "mom multitasking (boss level)" / "rawr (I'm hungry?)" / "yo earth snack detected" / "eye of judgment" / "important: Remember lunch" / "what is this even" / "2+2 = fish" / "definitely correct ✓" → text-as-style 的 production primitive · meta-creative ability (b) **多 zone text + 多个 mini doodle (恐龙 · 外星人 · pizza · UFO · cube · 食物 etc.)** 都准 · 跟 iPhone poster (15+ text · 分 zone) 同 pattern · 结构化布局赢 (c) **Identity 保持** + **scene 保留** · 同时 surreal 改造 · 是最难类 · 比 pathetic 维持像 pasta 物理感 + 跨 medium · 但 fidelity 仍高 Production: 个性化礼物 / SNS 头像 / 趣味海报 · viral 潜力高 (text 内容每次不同).
observed_once ¥0.5
view detail →
Pasta-Line Flat Lay Redraw Template
Pasta-Line Flat Lay Redraw Template
1:1
单 ref portrait → spaghetti pieces 拼出轮廓 · 跨 medium 物理材质 · sparse gap-heavy · pack avatar-redraw-2026q2 variant
项目内首验证 2026-05-01 · 57s wall · pack avatar-redraw-2026q2 variant 6/7. genuine pasta 形状 · pale yellow · uneven / bent / 不齐放置 · 大量空白 · 真"用面条拼脸"感. 跟 torn-paper 同 medium-transform 家族 · 但 pasta 是 sparse outline (vs paper 是 fill). Production: 餐饮 / 面食品牌 / 面条节 marketing / 食物纪录片 visual.
observed_once ¥0.5
view detail →
Whimsical Children's Book Redraw Template
Whimsical Children's Book Redraw Template
1:1
单 ref portrait → 童趣手绘绘本风 · 黑线轮廓 + 柔和 pastel 色块 + 抽象背景装饰 · 完整 DNA prompt · pack avatar-redraw-2026q2 variant
项目内首验证 2026-05-01 · 56s wall · pack avatar-redraw-2026q2 variant 7/7. 唯一**温柔系 redraw** (其他 5 是失败美学/极简) · 完整 DNA prompt (~1280 chars 中文 + EN keywords). 出片 cream 背景 · 软 pastel · 圆 sun/cloud/star · 心形装饰 · 鼠尾草绿 + 珊瑚橘. **Bonus: 模型自加 Hello Kitty patch 在 jacket** (auto-add IP · 童书风格联想). Production: 儿童礼物 / 母婴品牌 / 育儿 SNS / 个性化绘本 cover.
observed_once ¥0.5
view detail →
Pathetic Redraw Template
Pathetic Redraw Template
1:1
1 张真人照 + viral "pathetic / clumsy / MS Paint" prompt → 刻意烂的儿童画版你 · meme / 自嘲 / SNS 分享 · viral GPT Image 2 trend
项目内首验证 2026-05-01 · yuan_kid 4yo + viral "pathetic redraw" prompt. 第 3 次 polish ceiling test (SNAPSHOT_CANDID v1 + NYC vlogger v2 都 partial) · **这次 EXTREME anchor 真破了 polish ceiling** · 出图为刻意 sketchy 童画风格 (粗糙线条 · 头大 · 比例歪 · 但 identity 仍 vaguely 可辨). Hypothesis 升级 · polish ceiling 可破 · 但需要 insulting / extreme degradation anchor ('pathetic / clumsy / MS Paint / pixel-by-pixel' + 'draw it however you want' self-mocking closer) · 中性 'casual / amateur' anchor 不够. Sister · signature-snapshot-candid (现 3-sample · 1 break-through + 2 partial).
observed_once ¥0.5
view detail →
Business Headshot Template
Business Headshot Template
1:1
1 张真人照 + 模板 prompt → 商务化档案头像海报 · 身份保持 + 自动翻译 (中文姓名→EN) + 信息条 (姓名/职位/机构) · 1:1
项目内 3 sample 同照不同 placeholder 全 PASS (2026-05-01) · stability production-grade. Key insights · (a) title 暗示不渗透视觉 · 'Future Explorer'/'CEO Tencent' 仍商务装 · (b) identity preservation 即使 demographic mismatch (4yo+CEO) 也不 aged-up · 推 validated_2plus 仍需 cross-demographic (成年/不同人种). 原 id identity-transform-business-headshot (Option C refactor 2026-05-01 · 移到 templates 轴).
observed_once ¥0.5
prompt id: business_headshot_template_v1 view detail →
Cinematic Motion Shot Template
Cinematic Motion Shot Template
9:16
1 张真人照 + DSLR cinematic prompt → 9:16 全身动态镜头 · 身份保持 + 显式 negative prompt + face lock 直接指令 · golden hour 户外
项目内首验证 2026-05-01 · yuan_kid 4yo + viral cinematic motion prompt (9:16 全身 spinning fountain). 4 unique test 维度全 PASSED: (1) **9:16 aspect ApiMart · FIRST validation** · 直接 accept · 无需 fallback to 4:5 → unlocks 垂直 Reels/TikTok/Shorts 格式 on gpt-image-2 → 之前只验过 1:1/4:5/4:3/2:1 · 9:16 现入 valid set (2) **Negative prompt syntax · STRONG honor** · 'painting/illustration/CGI/plastic skin/...' 反向 anchor 全有效 → gpt-image-2 不像 SDXL 强 negative · 但 显式 'negative prompt:' prefix 有效降低相应 artifact → 新 prompt-engineering primitive · 适合 DSLR / photoreal 场景作 安全网 (3) **Face lock vs 'young woman' anchor 冲突 · FACE LOCK WON** · 4yo ref + 'young woman' subject anchor + 'do not alter face/hairstyle' directive · 结果: identity 保持 (East Asian girl · 略 age-compensate 至 ~7yo · 但仍是 child · NOT young woman) · OPPOSITE of porcelain test (where 'young woman' beat 'use for structure only') → **新 rule**: 显式 imperative ('do not alter face') 强于 descriptive constraint ('use for structure only') → directive 强度 > anchor 数量 (4) **Photoreal aesthetic** · DSLR-quality 完整达成 · golden hour rim glow + water droplets mid-air + 85mm bokeh + skin pores visible + 自然 motion blur on dress edges + wet fabric edges sticking Identity-preserving family · sister templates-business-headshot (1:1 商务化 · 同 'directive 直接指令' 范式) + sister templates-porcelain-sculpture-portrait (medium-transform · 反例参考).
observed_once ¥0.5
view detail →
High-Angle Fisheye Fashion Portrait Template
High-Angle Fisheye Fashion Portrait Template
1:1
pure T2I 5-item wardrobe + 高角度 fisheye + studio 渐变 → 杂志封面级 editorial streetwear · 无 ref · 1:1
项目内首验证 2026-05-01 · 5 unique 维度全 PASSED · 出片杂志封面级 (Vogue / Harper's Bazaar tier). 5 unique test 维度: (1) **Fisheye + high-angle 镜头语言** · 之前只测 85mm/wide/telephoto · 边缘 subtle 扭曲 + top-down perspective + 主体 looking up · gpt-image-2 理解 (2) **Editorial streetwear 美学** · 新 aesthetic anchor (vs photoreal/painting/sketch/ink wash) · magazine-cover 质感 (3) **Studio + 绿渐变背景** · 新背景类型 (vs outdoor/mural/landscape) · 干净 · soft diffused · 0 shadow (4) **5-item multi-layer wardrobe** · stress test · 全 honored · color-block puffer 3 色按 prompt 顺序成水平条带 (5) **Gesture spec** ('hand adjusting sunglasses') · model-pose archetype 完美执行 Default subject = East Asian young woman (无 ref · 模型默认 fashion editorial 选择). Production use: · 服装品牌 lookbook · 编辑造型 mockup · 个人 fashion 内容 · sneaker / hat / outerwear marketing. · 替换 wardrobe 5 元素 + 颜色描述即批量出新 look · template 化潜力高. Sister templates-cinematic-motion-shot (同 photoreal · 但有 ref + identity lock · 户外 vs 此条 studio). Sister templates-surreal-mural-portrait (同 pure T2I · 但 environmental composition vs 此条 portrait composition).
observed_once ¥0.5
view detail →
Function-Syntax Weighted 2×2 Creature Variations Template
Function-Syntax Weighted 2×2 Creature Variations Template
1:1
Function-style prompt + 数字权重 layer + 跨域 translation + meta-prompting → 2x2 grid 4 个 creature variants 带 per-cell label · 一次出 4 个 viral concept
项目内首验证 2026-05-01 · 单样本 · 7 unique 维度全 PASSED · 出片 Michelin-quality. 首样本: 4 extinct animals (Triceratops/T-Rex/Mammoth/Pteranodon) as sushi sculptures · 2x2 grid · 全标签. **3 个 framework-级新发现:** (a) **gpt-image-2 honor 显式 function-syntax prompt** · `Function NameXXX($Arg1, $Arg2)` + numbered steps + 权重 ::N + Negative ::-1 这套 LLM 风格 · 在 image gen 上有效 · 不是被忽略当 keyword bag · 启示: 复杂 multi-stage creative task 可以用 function 包装 vs flat description (b) **数字权重 ::N 作 priority hint 有效** · ::5 (Anchor) · ::4 (Morphology) · ::3 (Material) · ::2 (Light) · ::1 (Render) · ::-1 (Negative) · 不像 SDXL 那么强权重 · 但**层级感清晰** · 高权重 anchor 更显眼 · 启示: 跟 显式 'negative prompt:' (cinematic motion 验) 同属 prompt-engineering primitive 拓展 (c) **2x2 + 4 short labels (≤20 chars) ALL PASS** · 之前 Noor 'Noor 🌸' (5 chars) 经验上限上调 · 现在: "TYRANNOSAURUS REX" (17 chars) 全 readable + 拼写正确 · 但**仍是 grid 内多 short label · 不是 long-form text · 不要 extrapolate 到段落级** 其他 PASSED 维度: · Meta-prompting: model 自完成"分析 → 推断"两步 (auto-enumerate 4 iconic + auto-map 食材) · Cross-domain translation: animal anatomy ↔ sushi ingredients (mammoth fur → rice grains 完美) · 2x2 grid: cells 均匀 · production sweet spot (47s vs 5x5 14min) · Auto-enumeration: model 偏向 high-recognition subjects (跟 5x5 fox 同模式) Production use: · 餐饮品牌 marketing · 创意菜单 mock · viral SNS 4-cell 系列 · pre-viz 4 concept 看哪个最 viral · template 化: 替换 "extinct animals" + "sushi" 为其他 cross-domain pair (e.g., 神话动物 × 乐高 / 童话角色 × bento) Sister storyboard-5x5-style-grid (同 grid axis · 5x5 是 style 维 · 此条是 subject 维 4 variation). Sister templates-ink-wash-painting (同 placeholder template · 但 ink wash 单 subject · 此条 grid). Sister templates-surreal-mural-portrait (同 in-image text · 但 mural 是单 string · 此条 4 labels).
observed_once ¥0.5
view detail →
Real Human + Named Anime IP Photo Template
Real Human + Named Anime IP Photo Template
1:1
Real-human ref (face lock) + 命名 anime IP (e.g., Ninja Hattori) 同框 · model 智解 stylized 为 mascot suit performer · "我和 X" 周边/SNS · 1:1
项目内首验证 2026-05-01 · yuan_kid + Ninja Hattori + Kenichi-style 角色 · 7 维度 6 PASS + 1 partial · 40s wall. **3 个 framework-级新发现:** (a) **gpt-image-2 RECOGNIZES + REPRODUCES specific anime IP** (Ninja Hattori 准 · 不替换 generic ninja) · ApiMart endpoint image gen 政策宽松 (memory feedback_apimart_endpoints_audit_diff +1 验) · 但 IP 准确性有偏差 (Shinzo 被解为 Kenichi-style K-cap 角色 · model 知识局限) · production 含义: 经典 IP 周边可做 · 但商用须法务 (b) **CLEVER 解 photoreal-stylized 冲突** · 渲染 anime characters 为"主题乐园 mascot suit performer" · 不是 cel-shading overlay · 是真人在 cartoon 头套里 · 与 photoreal 真人共存合理 · 模型自创 "stylized" → "costumed performer" framework · 启示: anchor 冲突时 model 偏向 real-world plausible 解 (跟 'porcelain young woman' / cinematic motion 'face lock young woman' 同模式) (c) **3-character composition 稳定** · triangle 构图 · 比例正确 · pose coherent · 之前我们 0 测过 user + 2 character · 现在 validate 成立 其他 PASS: 100% face lock (yuan_kid 清晰 · 轻微 age comp ~5yo) · stylish modern outfit ✓ · 街景 + 自然光 ✓ · selfie pose ✓ Partial: "Zabardast" Hindi/Urdu 'fantastic' 大概率被忽略 · 用纯 EN/CN Production use: · 个人 + 经典动漫 IP 合影 · SNS / 周边 / 粉丝内容 · 主题乐园 mascot 互动模拟 · 跨 style "我和 X" 系列 (X 可换任何 cartoon IP) · 替换 ref + IP 即批量出新片 Sister templates-cinematic-motion-shot (同 ref + face lock · 但单人 · 此条 multi-character + cross-style). Sister templates-business-headshot (同 identity-preserving · 但单人 + 商务 · 此条娱乐). GOTCHA · IP 准确性 case-by-case 验 (Hattori 准 · Shinzo 不准 · 名 IP 越火越准). GOTCHA · 多语言 fashion vocab (Hindi/Urdu/etc.) 大概率忽略 · 用 EN/CN.
observed_once ¥0.5
view detail →
Pastel Cyberpunk Portrait · Anti-Neon Template
Pastel Cyberpunk Portrait · Anti-Neon Template
4:5
Anti-cyberpunk-cliché · 暖 pastel 调 + 显式 'no neon' negative + gallery aesthetic · 高端品牌 / 时尚 / 二次元 IP 现实化 · 4:5 portrait
项目内首验证 2026-05-02 · 28.9s wall · 5 维度全 PASSED · gallery-tier 出片 · 新 aesthetic 'Pastel Cyberpunk'. **核心 framework 贡献:** 验证 "explicit beats implicit" 规则跨 3 类 anchor 成立 (从此条进 atlas 第 3 条 sediment): · negative prompt overrides positive aesthetic (cinematic motion · 验) · color palette overrides genre default (此条 · 暖 pastel beats cyberpunk neon) · face lock directive overrides subject anchor (cinematic motion · 验) **5 维度执行:** · Anti-neon negative anchor → 0 neon · 0 glow · 仅自然光 · Warm pastel + cyberpunk fusion → cream + muted metallic + 粉 + soft blue · NEW family · School bus interior → 多排座椅 + luggage rack + 自然光 · 新环境 sediment · Gallery aesthetic → refined composition + DoF + 高端克制 · "Sleek + soft" 矛盾 anchor → 模型物理分离 (中 sleek + 侧 soft) · 创意空间分配解 **Production use:** · 高端时尚品牌 (anti-cliché 转 luxury restraint) · anime / 二次元 IP 现实化 · 教育 / 学院主题 with 未来感 · template-able (替换 character / environment / palette) Sister templates-cinematic-motion-shot (同 explicit-beats-implicit 范式 · 但 photoreal vs 此条 anime-inspired). Sister templates-fashion-fisheye-portrait (同 portrait + studio control · 但商业 vs 此条 gallery). Sister templates-real-plus-anime-ip-photo (同 anime-inspired aesthetic · 但 IP-character 同框 vs 此条单 character).
validated_2plus ¥0.5
view detail →
Selective Color Portrait with Heavy Brand Text Template
Selective Color Portrait with Heavy Brand Text Template
1:1
Selective color (face / frames in color · rest B&W) + 拥挤 brand-heavy 背景 + 6+ readable text strings + advanced "object-as-mask" 概念 · pure T2I · 1:1
项目内首验证 2026-05-01 · 36s wall · 7 维度全 PASSED · 出片 magazine 级. 首样本: punk teen + Nirvana tee + 2 photo frame (eye+lips 选色) + grunge wall (Ramones/Sex Pistols/PUNK NOT DEAD/RIOT). **4 个 framework-级新发现:** (a) **Selective color WORKS** (我们 0 测过) · 主要 frame 内色 · 但 model 智能延展到 brand items (Nirvana shirt + 红 RIOT 字) · 解 "rest is mostly B&W" 不是 "exclusively" · brand color hierarchy · 启示: 局部色 + 大面 desaturate 新可控 primitive (b) **In-image text 上限再上调 · 6 distinct strings** · 12-char ("PUNK NOT DEAD") 到 4-char ("RIOT") 全 readable · 6 strings 同框 · 上限 ladder: 5 (Noor) → 17 (sushi 单 label) → 6 strings × 4-12 chars (此条) · production: brand-heavy 拥挤场景完全可控 (c) **Music brand IPs 高准确度** · Nirvana smiley + Ramones + Sex Pistols 全准 · 比 anime IP (Hattori 准 / Shinzo 漂) 强 · music brand 训练密集 · 假设 IP 准确度: music > 顶 anime > 副 anime > 小众 (d) **"Photo frames as compositional mask"** · advanced concept 执行 · frame 不只 prop · 是 viewport / 局部色彩窗 · 可拓展: 镜子 / 窗外不同时间 / book pages 内异世界 其他 PASSED: editorial portrait · layered grunge wall · multi-prop (headphones/glasses/frames/tee/watch). Production use: · streetwear / punk / band marketing · 唱片店 / livehouse 内容 mock · "music tribute" 系列 (替换乐队 + 海报组) · selective-color 风格 photoshoot mock Sister templates-fashion-fisheye-portrait (同 photoreal pure T2I · studio vs 此条 grunge env). Sister templates-surreal-mural-portrait (同 in-image text · 单 string vs 此条 6 strings). Sister templates-function-weighted-2x2-creature (同 in-image text + brand · 但 grid · 此条单 portrait). GOTCHA: selective color 边界 model 自决 (不只 frame 内) · 严控需 "color ONLY in frames" 显式约束.
observed_once ¥0.5
view detail →
Lego Brick Concept Poster Template
Lego Brick Concept Poster Template
4:5
主题占位符 → 原创积木风格字体 + 概念视觉隐喻 + 多 zone 信息图 (字体样本 + 模块结构 + 颜色图例 + 细线标注) + 高级克制非玩具感 · 收藏级 design exhibition poster
项目内首验证 2026-05-01 · 主题=YUAN STUDIO · 46.5s wall · 8 维度全 PASSED · 出片 museum-quality. **brand-meta sample**: 用户首样本同时是我们 yuan.studio 的 brand poster · 收藏价值. 8 unique 维度全 PASS: (1) **Original brick-font generation** · 项目内首测 font construction · "YUAN STUDIO" 11 chars + 空格全可读 (2) **Concept exhibition poster 范式** · gallery-quality 排版 · grid 系统 · 干净留白 (3) **Multi-element infographic layout** · 6+ zones 全 honor (头标 + hero + 概念 + 字体样本 + 模块系统 + 颜色图例 + 应用场景 + 页脚) (4) **3D physical material** · 塑料 stud + 厚度 + 阴影 + 模块拼接 全可见 (5) **"不像儿童玩具广告" anti-anchor** · 完全 honor · 限定色调 · 杂志级 typo (6) **Auto visual metaphor** · 模型自创 "圆·元·源" 中文 wordplay (元 是 yuan 拆字 · brilliant!) + 圆环 brick 雕塑作 visual hook (7) **Auto aspect** · 4:5 declared · 海报正确比例 (8) **Brand-meta** · 自加 "© 2024 YUAN STUDIO" 页脚 + tagline "BUILDING IDEAS, BLOCK BY BLOCK" **Bonus shock findings:** · 完整 A-Z + 0-9 alphabet 都用 brick 渲染 (typeface specimen 36+ 字符) · 8 个 hex 色值 (model 自创 brand palette) · CN/EN 双语共存 (新木字体概念设计 / BUILDING IDEAS, BLOCK BY BLOCK) · text count 30+ distinct elements · 跟 iPhone poster 同 zoned-prompt-honor pattern Production use: · 任何品牌 launch 视觉 (替换 主题 即批量出 brand poster) · 字体 design 提案 · 设计展 / 概念展览物料 · 教育内容 (字体设计教学样本) Sister commercial-ui-ad-iphone-poster (同 长 structured prompt + multi-zone + 商业 poster · 但产品 hero · 此条 typeface-as-hero). Sister templates-function-weighted-2x2-creature (同 placeholder template + creative concept · 但 grid · 此条单 composition).
observed_once ¥0.5
view detail →
§11

Commercial UI Ad

商业 / 品牌 / 产品广告 · 双层视觉 (UI 冻结 + 内容运动) · 与 MV 同级新 vertical · formats/commercial_ui_ad/
1 item
iPhone Premium Campaign Poster (Commercial UI Ad)
iPhone Premium Campaign Poster (Commercial UI Ad)
4:5
创意-brief 长 prompt + multi-zone composition (top brand + mid hero + bottom feature strip + 背景 typography) + product hyper-detail · 4:5 OOH-ready · 此条**激活** commercial_ui_ad track
项目内首验证 2026-05-01 · 38.9s wall · 9 维度全 PASSED · 单样本 production-grade Apple-tier. **Track activation:** 此条作 commercial_ui_ad pack 首样本 (memory project_commercial_ui_ad_track 'scaffolded · pending' → 现 Gate 1 image PASS · Gate 2 i2v 待跑). **5 个 framework-级新发现 · 多个上限破:** (a) **超长 structured prompt 完全 honor** · 3356 chars · 12+ sections (FORMAT / CORE INTENT / SCENE / SUBJECT / PRODUCT / LIGHTING / COLOR / GRAPHIC / TYPOGRAPHY / FEATURE STRIP / COMPOSITION GRID / CAMERA SPECS) · 之前担心 prompt 过长会 degrade · 此条证伪 (无 quality drop · 38.9s wall 不长) · gpt-image-2 高 prompt 长度容忍 · 适合 creative-brief 型 prompt (b) **Text 上限 DRAMATIC revision** · 15+ distinct elements + 完整 body paragraph 全可读 · ladder: Noor 5 (2026-04) → sushi 17 单 (上午) → punk 6 strings (下午) → **此条 15+ elements + 25-词 paragraph (现在)** · 4 feature cards × 2 行各 (8 strings 全准) · Body paragraph "iPhone 17 Pro. Uncompromising power. A19 Pro chip. Advanced camera system. Designed for what's next." 全可读 · production: 商业 ad poster 文案多 layer 全可控 · 不需后期合成 (c) **Apple brand 最高准确度** · IP 准确度排序更新 Apple > music brand > 顶级 anime > 副 anime > 小众 (d) **Multi-zone single composition** validate (vs grid 范式) · 4 layered zones 同时 honor · 不退化 · 视觉 hierarchy 清晰 · zoned (cells unequal weight) vs grid (cells equal) (e) **iPhone 17 Pro 不存在但 model extrapolate** · Model 用 16-Pro 设计语言 + "17" 数字 + "Titanium Orange" 创意材质 · 合理外推 · 启示: future / fictional product 可用 brand 设计语言 + 显式 spec 锚定 其他 PASSED: · 模特 (clean editorial · neutral expression · 自然风发 · maroon jacket + off-white inner) ✓ · Hero pose (low angle · eye contact · phone forced perspective · forced foreground) ✓ · Glass morphism (pill + 4 cards 全玻璃感) ✓ · 三摄组 hyper-sharp + brushed titanium + lens reflection + Apple logo ✓ · Sparkle icons ✦ (3 围 phone) ✓ · Subtle grain · 暖 rim light · soft shadow ✓ Production use: · 商业 OOH / 数字 / SNS 海报 (4:5 declared global-ready) · 品牌 launch 视觉 mock · 产品发布会 visual treatment · 替换品牌 / 产品 / 模特即批量出 ad campaign · 4 feature cards = 4 卖点框架 · 任 product 可套 Next step (per formats/commercial_ui_ad/pipeline.md): Gate 2 = Seedance i2v 测此图作 ref · 看 UI-lock 能否 honor (背景动 · UI 冻). Sister templates-fashion-fisheye-portrait (同 photoreal pure T2I · 但 fashion 单 portrait vs 此条 multi-zone commercial) Sister templates-selective-color-portrait-with-text (同 heavy text · 但 6 vs 此条 15+ + paragraph) Sister templates-function-weighted-2x2-creature (同 structured prompt · 但 function syntax vs 此条 creative brief)
observed_once ¥0.5
view detail →