Skip to content

Spot it. Save it. Free, local-first Chrome extension for images, videos and prompts. Explore 2,000+ curated cases and connect your library to AI Agents via MCP. 提示词导演

15.4 KB
Raw Blame
import test from "node:test";
import assert from "node:assert/strict";

import {
  CLASSIFIER_VERSION,
  classifyContent,
  classifyCapturedContent,
  classifyImportedMedia,
  classifyImageCase,
  confirmClassification
} from "../extension/classifier.js";
import { CONTENT_IDS, CONTENT_ROLES, createContentType, createDefaultTaxonomy, removeContentType } from "../extension/taxonomy.js";

test("declared creative works use their medium even when prompt text resembles an article or a broad site rule", () => {
  const entry = { url: "https://example.com/work", title: "Workflow", text: "First use the attached image. Then follow these steps, because identity consistency matters.",
    sourceFacts: { pageType: "video", extractionMethod: "structured", originalPromptAvailable: true },
    mediaAssets: [{ id: "video", kind: "video", storageMode: "reference", sourceUrl: "https://cdn.example.com/movie.mp4" }] };
  const rules = [{ hostname: "example.com", pathIds: [CONTENT_IDS.tutorial], enabled: true }];
  assert.deepEqual(classifyContent(entry, rules).pathIds, [CONTENT_IDS.promptVideo]);
  assert.deepEqual(classifyContent({ ...entry, classification: { pathIds: [CONTENT_IDS.reference], status: "confirmed", source: "manual" } }, rules).pathIds, [CONTENT_IDS.reference]);
  assert.deepEqual(classifyContent({ ...entry, sourceFacts: {pageType: "article", extractionMethod: "structured"} }, rules).pathIds, [CONTENT_IDS.tutorial]);
});

test("universal prompt-structure teaching stays tutorial", () => {
  const result = classifyContent({
    text: "提示词结构教学:先定义主体,再补充环境、镜头和约束。这个框架适用于不同模型与题材。",
    title: "通用提示词写作方法"
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.tutorial]);
  assert.equal(result.status, "confirmed");
  assert.equal(result.source, "auto");
});

test("screenshot-only references use the fixed image-case type", () => {
  const result = classifyImageCase();
  assert.deepEqual(result.pathIds, [CONTENT_IDS.imageCase]);
  assert.equal(result.status, "confirmed");
  assert.equal(result.source, "manual");
});

test("generic classification keeps semantic web content behavior", () => {
  const pdf = classifyContent({
    title: "摄影资料.pdf",
    mediaAssets: [{ id: "pdf", kind: "document", mimeType: "application/pdf" }]
  });
  const note = classifyContent({ text: "下次测试低机位和更克制的色彩。", sourceKind: "quick_note" });
  const tutorial = classifyContent({
    title: "镜头教程.md",
    text: "镜头运动教学:第一步明确视线,第二步选择推镜,最后检查空间连续性。",
    mediaAssets: [{ id: "md", kind: "document", mimeType: "text/markdown" }]
  });

  assert.deepEqual(pdf.pathIds, [CONTENT_IDS.reference]);
  assert.deepEqual(note.pathIds, [CONTENT_IDS.reference]);
  assert.deepEqual(tutorial.pathIds, [CONTENT_IDS.tutorial]);
});

test("local imports use their file shape instead of guessing from the document body", () => {
  const tutorialDocument = classifyImportedMedia({
    title: "镜头教程.md",
    text: "教程:第一步明确视线,第二步选择推镜,最后检查空间连续性。",
    mediaAssets: [{ id: "md", kind: "document", mimeType: "text/markdown", storageMode: "managed" }]
  });
  const image = classifyImportedMedia({
    title: "参考图.png",
    text: "视频提示词教程",
    mediaAssets: [{ id: "image", kind: "image", mimeType: "image/png", storageMode: "managed" }]
  });
  const video = classifyImportedMedia({
    title: "镜头.mov",
    text: "图片提示词教程",
    mediaAssets: [{ id: "video", kind: "video", mimeType: "video/quicktime", storageMode: "managed" }]
  });
  const note = classifyImportedMedia({ text: "教程和提示词备忘", sourceKind: "quick_note" });

  assert.deepEqual(tutorialDocument.pathIds, [CONTENT_IDS.reference]);
  assert.deepEqual(image.pathIds, [CONTENT_IDS.imageCase]);
  assert.deepEqual(video.pathIds, [CONTENT_IDS.videoCase]);
  assert.deepEqual(note.pathIds, [CONTENT_IDS.reference]);
  assert.equal(tutorialDocument.source, "local_import");
});

test("workflow in a title does not turn an explicit image prompt into a tutorial", () => {
  const result = classifyContent({
    text: "Midjourney prompt: product hero image --ar 1:1",
    title: "Image workflow"
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.promptImage]);
});

test("a tutorial word in the title cannot override an explicit prompt body", () => {
  const result = classifyContent({
    text: "Midjourney prompt: editorial portrait --ar 4:5",
    title: "A complete tutorial collection"
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.promptImage]);
});

test("all universal tutorial media variants stay in the single tutorial type", () => {
  const video = classifyContent({
    text: "视频提示词教学:镜头运动应服务叙事节奏。先确定视线,再选择跟拍或推镜;这个方法适用于不同场景。"
  });
  const image = classifyContent({
    text: "生图教学:画幅决定视觉重心。人物肖像通常先确定构图,再选择光线;这些原则也适用于 Midjourney。"
  });
  const general = classifyContent({
    text: "提示词写作攻略:第一步明确主体,第二步补充限制条件,最后检查表达。"
  });

  assert.deepEqual(video.pathIds, [CONTENT_IDS.tutorial]);
  assert.deepEqual(image.pathIds, [CONTENT_IDS.tutorial]);
  assert.deepEqual(general.pathIds, [CONTENT_IDS.tutorial]);
});

test("a how-to wrapper around a concrete GPT Image prompt remains an image prompt", () => {
  const result = classifyContent({
    title: "How to turn an AI character into a real iPhone photo with GPT Image 2? Prompt below!",
    text: "Transform the person in the reference image into a real person. Create a casual iPhone photo with slight motion blur, imperfect framing, natural skin texture, and evening street light. Keep the same identity and clothes."
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.promptImage]);
});

test("a Midjourney style definition is tutorial metadata, not an image prompt", () => {
  const result = classifyContent({
    title: "Epic scale Midjourney style | Style library",
    text: "Epic scale is a visual style used to describe art that conveys grandeur and monumental scope. It is characterized by vast compositions, tiny figures, dramatic perspective, and mythic environments."
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.tutorial]);
});

test("a named style-catalog entry is tutorial even when its definition starts with Style colon", () => {
  const result = classifyContent({
    title: "Styles Buoy for an image model",
    text: "Castlevania Concept Art Style: A dark gothic fantasy concept-art style rendered with painterly linework, moody color, atmospheric fog, and dramatic vertical perspective."
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.tutorial]);
});

test("a style-catalog URL cannot override a concrete prompt body", () => {
  const result = classifyContent({
    title: "Midjourney prompt",
    url: "https://example.com/styles/editorial-portrait",
    text: "Create an editorial portrait with hard side light, restrained color, visible skin texture, and a quiet studio background --ar 4:5"
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.promptImage]);
});

test("a concrete Seedance generation specification remains a video prompt", () => {
  const result = classifyContent({
    title: "Seedance 2.0 prompt share",
    text: "Use the attached character sheet as the exact identity anchor. A single continuous handheld shot follows her through the market, then circles behind her as she runs. Wind moves her coat and the camera pulls back for the final reveal."
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.promptVideo]);
});

test("a Style field inside a concrete video specification is not mistaken for a style tutorial", () => {
  const result = classifyContent({
    title: "Video prompt",
    text: "Style: A hand-painted cinematic look. A single continuous handheld shot follows the warrior, then the camera circles behind her and pulls back for the final reveal."
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.promptVideo]);
});

test("Japanese Seedance video specifications use the video prior without making it absolute", () => {
  const result = classifyContent({
    title: "Seedance 2.0 動画プロンプト共有",
    text: "使用参照: @image1。15秒、1:1。高密度な劇場版2Dセルアニメーション。映像では同じ主人公を維持し、カメラが人物を追い、最後に上空へ引く。"
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.promptVideo]);
});

test("Seedance workflow teaching with transferable reasoning is tutorial", () => {
  const result = classifyContent({
    title: "Seedance camera movement tutorial",
    text: "Why word choice matters: camera verbs define spatial direction, while subject verbs define action. Start with the visual anchor, then describe one movement at a time. This structure helps preserve continuity across different scenes."
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.tutorial]);
});

test("a cross-model consistency workflow is tutorial", () => {
  const result = classifyContent({
    title: "GPT Image 2 + Seedance 2 workflow",
    text: "GPT Image builds the character and Seedance brings it to life. Most creators treat these as separate steps, which causes identity drift. The reusable method is to create a character sheet first, then use it as the identity anchor for each shot."
  });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.tutorial]);
});

test("classifies timeline video prompts and explicit image prompts", () => {
  const video = classifyContent({
    text: "0-4s: camera tracks the hero. 4-8s: slow dolly out.",
    title: "Seedance prompt"
  });
  const image = classifyContent({
    text: "Flux image prompt, editorial product photography --ar 4:5"
  });

  assert.deepEqual(video.pathIds, [CONTENT_IDS.promptVideo]);
  assert.deepEqual(image.pathIds, [CONTENT_IDS.promptImage]);
});

test("leaves unclear content waiting for confirmation instead of guessing", () => {
  const result = classifyContent({ text: "Make it beautiful and memorable." });

  assert.deepEqual(result.pathIds, []);
  assert.equal(result.status, "needs_review");
  assert.equal(result.classifierVersion, CLASSIFIER_VERSION);
});

test("a short ambiguous caption with a saved screenshot defaults to an editable image case", () => {
  const result = classifyContent({ text: "A strange style I want to remember.", hasScreenshot: true });

  assert.deepEqual(result.pathIds, [CONTENT_IDS.imageCase]);
  assert.equal(result.status, "confirmed");
  assert.equal(result.source, "auto");
});

test("manual confirmation survives future automatic classification", () => {
  const manual = confirmClassification(
    { text: "0-4s: camera tracks the hero" },
    [CONTENT_IDS.promptImage]
  );
  const recalculated = classifyContent(manual);

  assert.deepEqual(recalculated.pathIds, [CONTENT_IDS.promptImage]);
  assert.equal(recalculated.source, "manual");
});

test("an explicit hostname rule is applied before local classification", () => {
  const result = classifyContent(
    {
      text: "An otherwise unclear saved example",
      url: "https://community.example.com/project/12"
    },
    [{ hostname: "community.example.com", pathIds: [CONTENT_IDS.promptVideo] }]
  );

  assert.deepEqual(result.pathIds, [CONTENT_IDS.promptVideo]);
  assert.equal(result.source, "source_rule");
});

test("automatic classification targets a user category by stable creative role", () => {
  let taxonomy = removeContentType(createDefaultTaxonomy(), CONTENT_IDS.promptImage);
  taxonomy = createContentType(taxonomy, {
    id: "content:campaign-prompts",
    name: "广告生图稿",
    role: CONTENT_ROLES.promptImage
  });
  const result = classifyContent({ text: "Midjourney prompt: editorial product image --ar 4:5" }, [], taxonomy);
  assert.deepEqual(result.pathIds, ["content:campaign-prompts"]);
});


test("capture defaults use selected media, with video taking priority over prose and images", () => {
  const image = { id: "image", kind: "image" };
  const video = { id: "video", kind: "video" };
  const entry = { text: "图片构图参考:海面,薄雾,孤舟。", mediaAssets: [image, video] };
  assert.deepEqual(classifyCapturedContent(entry).pathIds, [CONTENT_IDS.promptVideo]);
  assert.deepEqual(classifyCapturedContent({ ...entry, mediaAssets: [image] }).pathIds, [CONTENT_IDS.promptImage]);
  assert.deepEqual(classifyCapturedContent({ ...entry, text: "" }).pathIds, [CONTENT_IDS.videoCase]);
  assert.deepEqual(classifyCapturedContent({ ...entry, text: "", mediaAssets: [image] }).pathIds, [CONTENT_IDS.imageCase]);
});

test("document capture stays reference even with tutorial text and video illustrations", () => {
  const entry = { text: "教程:第一步明确主体,第二步检查构图,最后说明为什么这个方法适用于不同场景。",
    sourceFacts: { pageType: "article", extractionMethod: "dom" }, mediaAssets: [{ id: "v", kind: "video" }] };
  assert.deepEqual(classifyCapturedContent(entry).pathIds, [CONTENT_IDS.reference]);
  assert.deepEqual(classifyCapturedContent(confirmClassification(entry, [CONTENT_IDS.promptImage])).pathIds, [CONTENT_IDS.promptImage]);
});

test("capture respects missing prompt evidence, but per-media original prompts remain usable", () => {
  const entry = { text: "作品宣传介绍", sourceFacts: { pageType: "artwork", originalPromptAvailable: false },
    mediaAssets: [{ id: "v", kind: "video" }, { id: "cover", kind: "image", usage: "poster" }] };
  assert.deepEqual(classifyCapturedContent(entry).pathIds, [CONTENT_IDS.videoCase]);
  assert.deepEqual(classifyCapturedContent({ ...entry, mediaPrompts: [{ assetId: "v", source: "webpage", text: "镜头跟随人物。" }] }).pathIds, [CONTENT_IDS.promptVideo]);
  assert.deepEqual(classifyCapturedContent({ ...entry, mediaAssets: [entry.mediaAssets[1]], text: "" }).pathIds, []);
});


test("platform defaults agree across capture and later reclassification, regardless of introductory prose", () => {
  for (const classify of [classifyContent, classifyCapturedContent]) {
    for (const host of ["www.youtube.com", "youtu.be", "www.bilibili.com", "www.douyin.com", "www.tiktok.com"]) {
      assert.deepEqual(classify({ url: `https://${host}/video/123`, text: "Prompt: create a cinematic video. 教程简介" }).pathIds, [CONTENT_IDS.videoCase], host);
    }
    for (const host of ["www.pinterest.com", "www.artstation.com", "www.behance.net"]) {
      assert.deepEqual(classify({ url: `https://${host}/work/123`, text: "Prompt: portrait --ar 4:5" }).pathIds, [CONTENT_IDS.imageCase], host);
    }
    assert.deepEqual(classify({ url: "https://higgsfield.ai/asset/123", mediaAssets: [{ id: "v", kind: "video" }], sourceFacts: { originalPromptAvailable: false } }).pathIds, [CONTENT_IDS.promptVideo]);
    assert.deepEqual(classify({ url: "https://www.midjourney.com/jobs/123" }).pathIds, [CONTENT_IDS.promptImage]);
    assert.deepEqual(classify({ url: "https://pinterest.com/pin/123", mediaAssets: [{ id: "v", kind: "video" }] }).pathIds, [CONTENT_IDS.videoCase]);
    assert.deepEqual(classify({ url: "https://youtube.com.evil.example/watch" }).pathIds, []);
    const entry = { url: "https://www.bilibili.com/video/123" };
    assert.deepEqual(classify(confirmClassification(entry, [CONTENT_IDS.promptImage])).pathIds, [CONTENT_IDS.promptImage]);
    assert.deepEqual(classify(entry, [{ hostname: "www.bilibili.com", pathIds: [CONTENT_IDS.reference] }]).pathIds, [CONTENT_IDS.reference]);
  }
});