Back to Details

amz-ai-image-policy-compliance

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

AI Image Policy Compliance

AI图片政策合规审核

AI image generation became mainstream for Amazon sellers in 2025. Then the 2026 disclosure and representation policy raised the bar. Two rules sit at the core: sellers must clearly disclose images that are substantially AI-generated, and every image must accurately represent the physical product that ships, regardless of the tool used to make it. Get either wrong and the image can be suppressed, sometimes with a misrepresentation flag on the account. A 5-gate audit before submission catches the issues before upload.
2025年，AI图片生成在亚马逊卖家中成为主流。随后出台的2026年披露与陈述政策提高了要求。核心规则有两条：卖家必须明确披露经过大量AI生成的图片；无论使用何种工具制作，每张图片都必须准确展示实际发货的实物商品。违反任意一条都可能导致图片被屏蔽，有时还会给账户打上虚假陈述标记。提交前通过五关卡审核可在上传前发现问题。

When to use this

适用场景

  • Generated a lifestyle scene with Nano Banana or Gemini and want to ship it
  • Replaced a real product photo with a CGI/AI render
  • A composite where the product is real but the model/scene is AI
  • Pre-launch image set review before bulk upload
  • 使用Nano Banana或Gemini生成了生活场景图并打算发布
  • 用CGI/AI渲染图替换了真实商品照片
  • 商品为真实拍摄但场景是AI生成的合成图
  • 批量上传前对预发布图片集进行审核

The framework. The Five-Gate AI Image Audit

框架：五关卡AI图片审核流程

Each image passes through 5 gates in order. Fail any gate, fix before next.
GateQuestionPass condition
1. Product photo authenticIs the product depiction a real photograph or AI?Real photo of actual unit. an AI-fabricated product depiction risks the misrepresentation rule.
2. Scene authenticityIf the scene is AI, does it accurately depict the product?Color, dimensions, texture, materials match the physical product. tight match, this skill uses a tolerance heuristic, not an Amazon number
3. Accuracy vs referenceCompared to real product photo, what is the delta?No invented features, no exaggerated size, no false materials
4. DisclosureIs the image substantially AI-generated, and if so is it disclosed?If substantially AI-generated, a clear plain-language disclosure is added prominently (Amazon points to the product description). minor retouching may not require it
5. Policy risk flagsDoes the image imply something it should not?No implied human endorsement, no faked use cases, no implied performance claims, no faked before/after
The Gate 2 match thresholds this skill applies, a color difference under roughly Delta E 5 and a size deviation under about 5 percent, are the skill's own QA heuristic for flagging drift. Amazon does not publish a numeric tolerance. its actual rule is qualitative: disclose substantial AI generation and represent the real product accurately.
每张图片需依次通过5个关卡，任意一关不通过需修复后再进入下一关。
关卡问题通过条件
1. 商品照片真实性商品展示是真实照片还是AI生成的？真实拍摄的实物照片。AI虚构的商品展示会违反虚假陈述规则。
2. 场景真实性如果场景是AI生成的，是否准确展示了商品？颜色、尺寸、纹理、材质与实物商品匹配。需高度匹配，本技能采用容差启发式判断，无亚马逊官方数值标准
3. 与参考图的一致性与真实商品照片相比，差异有多大？无虚构功能、无尺寸夸大、无虚假材质
4. 披露要求图片是否为大量AI生成？若是，是否已披露？若为大量AI生成，需在显眼位置添加清晰易懂的披露说明（亚马逊建议放在商品描述中）。轻微修图可能无需披露
5. 政策风险标记图片是否存在不当暗示？无隐含的真人背书、无伪造使用场景、无隐含性能宣称、无伪造前后对比图
本技能在关卡2中采用的匹配阈值：颜色差异大致低于Delta E 5，尺寸偏差低于约5%，这是本技能用于识别偏差的QA启发式标准，亚马逊并未公布数值容差。其实际规则为定性要求：披露大量AI生成内容，并准确展示真实商品。

Step by step

操作步骤

  1. Inventory the image set. For each image, classify: 100% real photo, 100% AI, or hybrid (real product + AI scene).
  2. Gate 1: Product photo authenticity. For each image, confirm the product pixels are from a real photograph of the actual unit. an AI-fabricated product depiction runs straight into the 2026 misrepresentation rule.
  3. Gate 2: Scene authenticity. If the scene is AI, hold up a real reference photo of the product. Check color, dimensions, visible materials and textures against it. The thresholds this skill uses below (a Delta E color limit, a percentage size band) are the skill's own QA heuristic for catching drift, not an Amazon-published rule. Mismatch = revise prompt.
  4. Gate 3: Accuracy vs reference. Visual compare to listing main image. Flag any feature that exists in the AI image but not the product (extra button, fake LED, fake texture). Strip the invention.
  5. Gate 4: Disclosure. If the image is substantially AI-generated, add a clear, plain-language disclosure that buyers can understand. Amazon's current guidance points to the product description, near the top. Minor retouching (crop, brightness, background removal) generally does not require disclosure. confirm the current placement and wording in Amazon's policy, since this is tightening through 2026.
  6. Gate 5: Policy risk flags. Run the policy checklist below. Common triggers: humans appearing to endorse the product, before/after weight loss scenes, food preparation showing impossible results, kids using adult products.
  7. Final pass/fail per image. All 5 gates pass = ship. Any fail = revise.
  8. Run the quality check, then submit.
  1. 盘点图片集：对每张图片进行分类：100%真实照片、100%AI生成、混合图（真实商品+AI场景）。
  2. 关卡1：商品照片真实性：确认每张图片中的商品像素来自真实实物的拍摄照片。AI虚构的商品展示直接违反2026年虚假陈述规则。
  3. 关卡2：场景真实性：如果场景是AI生成的，拿出商品的真实参考照片，对比颜色、尺寸、可见材质和纹理。以下为本技能使用的阈值（Delta E颜色限制、百分比尺寸范围），是本技能用于识别偏差的QA启发式标准，并非亚马逊公布的规则。不匹配则需修改提示词。
  4. 关卡3：与参考图的一致性：与listing主图进行视觉对比，标记AI图片中存在但实物没有的特征（额外按钮、虚假LED、虚假纹理），移除这些虚构内容。
  5. 关卡4：披露要求：若图片为大量AI生成，添加清晰易懂的披露说明，让买家能够理解。亚马逊当前指引建议放在商品描述顶部。轻微修图（裁剪、亮度调整、背景移除）通常无需披露。发布前需确认亚马逊政策中当前的放置位置和措辞要求，因为2026年期间规则会不断收紧。
  6. 关卡5：政策风险标记：对照下方的政策清单检查。常见触发点：看似为商品背书的人物、减肥前后对比场景、展示不可能结果的食品制作过程、儿童使用成人产品。
  7. 单张图片最终审核结果：全部5关通过即可发布；任意一关不通过则需修改。
  8. 进行质量检查，然后提交。

Output format

输出格式

undefined
undefined

AI Image Audit. [SKU / image set]

AI Image Audit. [SKU / image set]

Per-image pass/fail Image # | Gate 1 | Gate 2 | Gate 3 | Gate 4 | Gate 5 | Verdict | Fix needed 1 | PASS | PASS | FAIL | PASS | PASS | REVISE | Remove fake LED indicator 2 | PASS | PASS | PASS | PASS | PASS | SHIP | 3 | FAIL | - | - | - | - | REJECT | Product is fully AI, re-shoot real
Disclosure to apply
  • Per image, classify: AI-generated [yes/no], Composite [yes/no]
  • If any shipping image is substantially AI-generated, add a clear plain-language disclosure prominently (Amazon's current guidance points to the product description). confirm the exact placement and wording in current Amazon policy before publishing.
Policy risk summary [Any Gate 5 flags surfaced with specific revision instructions]
undefined
Per-image pass/fail Image # | Gate 1 | Gate 2 | Gate 3 | Gate 4 | Gate 5 | Verdict | Fix needed 1 | PASS | PASS | FAIL | PASS | PASS | REVISE | Remove fake LED indicator 2 | PASS | PASS | PASS | PASS | PASS | SHIP | 3 | FAIL | - | - | - | - | REJECT | Product is fully AI, re-shoot real
Disclosure to apply
  • Per image, classify: AI-generated [yes/no], Composite [yes/no]
  • If any shipping image is substantially AI-generated, add a clear plain-language disclosure prominently (Amazon's current guidance points to the product description). confirm the exact placement and wording in current Amazon policy before publishing.
Policy risk summary [Any Gate 5 flags surfaced with specific revision instructions]
undefined

Worked example

示例案例

8-image set for kitchen knife block. Real product photographed against white. Lifestyle scenes (5 of 8) generated in Nano Banana with the real product composited in.
  • Images 1-3: pure product on white, real photos. All gates PASS. SHIP.
  • Image 4: AI kitchen scene, knife block composited. Gate 1 PASS, Gate 2 FAIL (wood handle color is #6B3410 in AI but actual product is #4A2510). Revise prompt with hex code. Resubmit.
  • Image 5: AI scene with hand using knife. Gate 5 FAIL (implied human use endorsement, hand is AI-generated). Revise to remove hand, show knife in cutting board mid-scene without user.
  • Image 6: AI before/after of dull vs sharp blade cut. Gate 5 FAIL (performance claim implied). Replace with neutral product detail shot.
  • Images 7-8: AI scenes, all gates pass. SHIP.
Result: 5 of 8 ship as-is, 3 revised, 0 rejected by Amazon on upload. The three caught here (a color mismatch, an implied-endorsement hand, a faked before/after) are exactly the kind of issue that draws suppression or a misrepresentation flag when it slips through.
某厨房刀架的8张图片集：真实商品在白色背景下拍摄，5张生活场景图由Nano Banana生成并合成了真实商品。
  • 图片1-3：纯商品白底图，真实照片。所有关卡通过，可发布。
  • 图片4：AI厨房场景，合成刀架。关卡1通过，关卡2不通过（AI图中木质手柄颜色为#6B3410，而实物为#4A2510）。需修改提示词并加入十六进制颜色码，重新提交。
  • 图片5：AI场景中手使用刀具。关卡5不通过（隐含真人使用背书，手部为AI生成）。需修改移除手部，展示刀具在砧板上的场景，无使用者。
  • 图片6：AI生成的刀刃钝锐切割前后对比图。关卡5不通过（隐含性能宣称）。替换为中性的商品细节图。
  • 图片7-8：AI场景图，所有关卡通过，可发布。
结果：8张图片中5张可直接发布，3张需修改，0张在上传时被亚马逊拒绝。这里发现的三个问题（颜色不匹配、隐含背书的手部、伪造前后对比图）正是会导致图片被屏蔽或账户被标记虚假陈述的典型问题。

Quality check

质量检查

  • Real reference photo used for Gate 2 comparison, not a render
  • Color and size check actually performed against the reference, not eyeballed (using the skill's heuristic, not an Amazon number)
  • Disclosure added for any substantially AI-generated image, per current Amazon placement guidance
  • Gate 5 checklist reviewed against current Amazon 2026 policy doc
  • Each image classified AI-generated yes/no for internal tracking
  • 关卡2对比使用真实参考照片，而非渲染图
  • 颜色和尺寸检查需实际对照参考图，而非仅凭目测（使用本技能的启发式标准，而非亚马逊官方数值）
  • 所有大量AI生成的图片均已按亚马逊当前放置指引添加披露说明
  • 关卡5清单已对照亚马逊2026年最新政策文档审核
  • 每张图片已标记是否为AI生成，用于内部追踪

Common mistakes

常见错误

  • Fabricating the product itself with AI. A made-up product depiction collides with the 2026 misrepresentation rule. show the real unit.
  • Skipping color check. Customers return when the real product is darker/lighter than the AI scene.
  • Skipping disclosure. A substantially AI-generated image needs a clear disclosure. confirm Amazon's current placement, do not assume image alt text alone covers it.
  • AI humans implying endorsement. Hands holding the product, models wearing apparel. All Gate 5 triggers.
  • Fake before/after scenes. Weight loss, hair growth, cleaning results. Instant policy violation.
  • 用AI虚构商品本身：虚构的商品展示违反2026年虚假陈述规则，需展示真实实物。
  • 跳过颜色检查：当实物与AI场景图颜色深浅不符时，顾客会退货。
  • 跳过披露步骤：大量AI生成的图片需添加清晰披露说明。需确认亚马逊当前的放置要求，不要假设仅图片替代文本即可满足要求。
  • AI人物隐含背书：手持商品的手、穿着服饰的模特，均为关卡5触发点。
  • 伪造前后对比场景：减肥、生发、清洁效果等，均属直接违反政策。

Built by Jay GPT Pro

Built by Jay GPT Pro

Part of Amazon Pro Skills. Production-grade skills for serious Amazon sellers. Free and open. Built by Jay Margaliot.
I share a new AI play for Amazon sellers every week, free, in my WhatsApp group. Join here: https://chat.whatsapp.com/ILX65p1yWcaIG3c9WGHpTY
Part of Amazon Pro Skills. Production-grade skills for serious Amazon sellers. Free and open. Built by Jay Margaliot.
I share a new AI play for Amazon sellers every week, free, in my WhatsApp group. Join here: https://chat.whatsapp.com/ILX65p1yWcaIG3c9WGHpTY