openclaw-voc-skill/claude-code/fmode-image-set/scripts/test-plan-fixture.mjs

import { compileProductLock } from "../mcp/src/providers/commerce-vision-provider.mjs";

function repeatToLength(seed, minimum) {
  let result = seed;
  while ([...result].length < minimum) result += ` ${seed}`;
  return result;
}

const identityManifest = {
  must_preserve: ["完整外形轮廓", "商品数量", "主体颜色", "可见结构", "标识位置"],
  must_not_add: ["额外配件", "额外文字", "不存在的纽扣"],
  unknown_do_not_assert: ["内部结构", "真实尺寸", "材质成分"],
};
identityManifest.product_lock = compileProductLock(identityManifest);

const slotConfigs = [
  ["hero", "isolated_product", "simple", "看清完整商品外观与数量", ["visible_fact:1"], "", "纯净无缝背景中只放置一个完整商品，保留充分边距、自然接触阴影与清楚轮廓。"],
  ["selling_point", "material_demonstration", "medium", "理解第一项可见材质价值", ["visible-sp-1"], "visible-sp-1", "以单一方向的掠射光划过商品表面，并用明暗渐变和局部微观纹理展示第一项可见材质特征；画面仍保留足够完整外形作为识别锚点，不加入说明文字或虚构性能。"],
  ["scene", "environmental_scene", "complex", "理解商品与完整生活空间的关系", ["suggested_scene:1"], "", "建立具有前景家具、中景商品与远景建筑层次的宽幅生活空间。商品自然位于真实承托面上，附近道具只用于交代尺度与情境，左右留出呼吸空间；环境细节、材质反射与光影方向互相呼应，不宣称未经证实的适用性。"],
  ["selling_point", "feature_demonstration", "medium", "理解第二项可见结构价值", ["visible-sp-2"], "visible-sp-2", "用侧向中近景展示与第二项卖点相关的可见结构，并让完整商品处于同一连续空间的后景作为关系参照；通过景深和连续空间层次区分重点，禁止虚构内部构造、功能或性能。"],
  ["decision_support", "multi_angle", "medium", "从多个真实可见角度完成外观判断", ["visible_fact:2"], "", "在唯一一张多格构图中，以等比例、等光色的三个外观视角建立横向阅读顺序；所有视角都只展示参考图支持的外部表面，间距清楚，不生成未知背面、内部、附件或说明文字。"],
  ["scene", "human_interaction", "complex", "理解人物动作与商品之间的关系", ["suggested_audience:1"], "", "人物从画面侧边进入，以明确、自然且符合商品可见结构的手部动作与商品发生联系。前景动作、中景商品和背景环境形成三层空间，服装、道具与光线克制统一；商品仍是视觉焦点，不遮挡身份特征，不添加未证实用途。"],
  ["decision_support", "usage_sequence", "medium", "理解商品在单一使用时刻中的状态关系", ["visible_fact:3"], "", "在一个连续、真实的桌面或生活空间中呈现商品与必要使用动作的前后关系，以视线方向、手部位置和物体间距传达顺序，保持单一连续画面，不加入步骤文字、测量值、营销结论或未知附件。"],
].map(([category, render, complexity, decision, evidence, sellingPointId, scene]) => ({
  category, render, complexity, decision, evidence, sellingPointId, scene,
}));

function buildModules(index, config) {
  const sceneMinimum = config.complexity === "complex" ? 175 : config.complexity === "medium" ? 75 : 25;
  return {
    product_lock: identityManifest.product_lock,
    image_job: `第${index + 1}张图只承担一个决策任务：${config.decision}。`,
    scene: repeatToLength(config.scene, sceneMinimum),
    composition: `第${index + 1}张使用独立的裁切、占比、边距、空间深度与阅读顺序，不能复用其他槽位的机位关系。`,
    lighting_material: `第${index + 1}张使用独立光线方向、光源面积、软硬度、色温、阴影和真实材质响应。`,
    camera_style: config.complexity === "complex"
      ? `为第${index + 1}张指定独立镜头焦段、焦平面和景深过渡。`
      : `第${index + 1}张采用独立的商业摄影机位与成像方式。`,
    text_layout: "",
    negative_constraints: `禁止商品漂移、无关物体、重复商品、虚构卖点、水印和额外品牌；禁止断言内部结构、真实尺寸或材质成分。`,
  };
}

export function compilePrompt(modules) {
  return Object.values(modules).filter(Boolean).join(" ");
}

export function buildValidPlan(
  category = "generic consumer product",
  reference = "https://images.example.test/product.jpg",
  runId = `cross-category-${category.replace(/\s+/g, "-")}`,
) {
  return {
    schema_version: "0.6",
    run_id: runId,
    status: "ready",
    user_output_mode: "images_only",
    request: {
      instruction: "这是我的产品，帮我生成一套通用电商七图。",
      reference_images: [reference],
    },
    context: {
      platform: "general_ecommerce",
      market: "unspecified",
      voc: { status: "unavailable", findings: [] },
    },
    product_understanding: {
      category,
      product_name: { value: category, basis: "verified visible evidence", confidence: "medium" },
      visible_facts: ["完整外形可见", "正面与侧面轮廓可见", "可见操作结构", "主色和表面材质观感可见"],
      verified_visible_facts: ["完整外形可见", "正面与侧面轮廓可见", "可见操作结构", "主色和表面材质观感可见"],
      inferred_attributes: [],
      unknown_features: ["内部结构", "真实尺寸", "材质成分"],
      identity_manifest: identityManifest,
      provided_facts: [],
      candidate_selling_points: [
        { id: "visible-sp-1", value: "可见材质与主色形成清晰的视觉识别", basis_refs: ["visible_fact:4"], confidence: "high" },
        { id: "visible-sp-2", value: "可见结构关系让商品外观层次容易理解", basis_refs: ["visible_fact:2", "visible_fact:3"], confidence: "high" },
      ],
      visible_specs: { visible_quantity: "1", exterior_state: "complete" },
      creative_positioning: ["克制、清晰的商业视觉"],
      suggested_audiences: ["重视清晰商品信息的消费者"],
      suggested_scenes: ["整洁、可信的日常使用环境"],
      unsupported_claims: ["经过认证", "超长耐用"],
      unknowns: ["内部结构", "真实尺寸", "材质成分"],
    },
    global_consistency_lock: {
      product_invariants: identityManifest.must_preserve,
      visual_system: ["统一商品身份与真实材质响应"],
      prohibited_changes: identityManifest.must_not_add,
    },
    set_blueprint: {
      business_allocation: { hero: 1, scene: 2, selling_point: 2, decision_support: 2 },
      visual_thesis: "用七个不可互换的购买任务让商品可识别、可理解、可代入并可判断。",
      narrative_arc: ["识别", "理解价值", "代入", "决策"],
      shared_visual_dna: ["商品身份", "核心色彩", "真实材质"],
      intentional_variations: ["业务任务", "空间", "人物", "景别", "信息关系"],
    },
    images: slotConfigs.map((config, index) => {
      const modules = buildModules(index, config);
      return {
        id: String(index + 1).padStart(2, "0"),
        role: config.category,
        business_category: config.category,
        decision_subject: config.decision,
        evidence_refs: config.evidence,
        selling_point_id: config.sellingPointId,
        purpose: config.decision,
        voc_refs: [],
        set_design: {
          render_family: config.render,
          scene_complexity: config.complexity,
          unique_visual_hook: `槽位${index + 1}使用${config.render}形成不可互换的视觉任务。`,
          contrast_with_previous: index === 0 ? "开场" : `槽位${index + 1}改变景别、空间、动作和信息关系。`,
        },
        shot_spec: {
          narrative_job: `job-${index + 1}`,
          composition_family: ["centered hero", "diagonal material field", "wide room depth", "single-space feature relation", "multi-angle grid", "human action frame", "continuous usage moment"][index],
          shot_scale: ["full product", "medium product", "wide environment", "detail and whole", "multi-view", "human medium-wide", "context close-medium"][index],
          camera_angle: ["eye level", "low oblique", "high wide", "side close", "three-quarter grid", "shoulder height", "top-oblique context"][index],
          camera_axis: ["front", "low", "three_quarter_left", "side", "three_quarter_right", "front", "top"][index],
          framing_mode: ["full_product", "detail", "product_in_context", "full_product", "multi_panel", "human_action", "product_in_context"][index],
          product_pose: ["upright", "detail_crop", "upright", "arranged_sequence", "multi_view", "held", "laid_flat"][index],
          product_placement: `distinct placement ${index + 1}`,
          product_occupancy_pct: [70, 45, 25, 50, 60, 20, 55][index],
          depth_structure: `distinct depth structure ${index + 1}`,
          background_family: ["pure seamless", "material field", "lived-in room", "feature stage", "neutral grid", "human environment", "continuous context"][index],
          lighting_family: ["even studio", "directional material", "window ambient", "edge definition", "flat comparative", "motivated environmental", "clear contextual"][index],
          interaction_mode: ["single accurate product", "material relation", "environment placement", "feature relation", "multiple exterior views", "human action", "single moment relation"][index],
          information_density: ["low", "medium", "medium", "medium", "high", "medium", "medium"][index],
        },
        prompt_modules: modules,
        generation_prompt: compilePrompt(modules),
        negative_prompt: "deformation, watermark",
        overlay_copy: [],
        reference_images: [reference],
        model_parameters: { adapter: "jimeng_img_v4", width: 2048, height: 2048, scale: 0.6, force_single: true },
        acceptance_checks: ["product identity preserved", "decision task is visible"],
      };
    }),
  };
}