Apple is embracing the fantasy of AI photo editing

苹果正在拥抱 AI 照片编辑的“幻想”

The company has some new ideas on ‘What is a photo?’ 这家公司对于“什么是照片”有了新的见解。

Apple used to question whether generative AI-powered editing features were worth the risk of distorting our perceptions of the world. Now it seems Apple no longer believes that photos should accurately capture reality. At WWDC 2026, the company announced a host of new AI-powered photo editing tools. They give users effortless powers of manipulating images that Apple still refers to as “photos.” 苹果曾经质疑，由生成式 AI 驱动的编辑功能是否值得冒着扭曲我们世界认知的风险。现在看来，苹果不再认为照片必须准确地捕捉现实。在 WWDC 2026 大会上，该公司宣布了一系列新的 AI 照片编辑工具。这些工具赋予了用户轻松操纵图像的能力，而苹果依然将这些图像称为“照片”。

Two years ago, Apple launched Clean Up — an AI-powered object removal tool in Apple’s Photo app that’s similar to the Magic Eraser feature in Google Photos. At the time, Apple software chief Craig Federighi said that it was important for the company to “purvey accurate information, not fantasy.” The company seemed hesitant to provide more extensive AI editing tools, while Google and Samsung charged ahead with editing suites that allow you to add almost anything to photographs by just describing it — including explosions, drug paraphernalia, and other potentially harmful inclusions. 两年前，苹果推出了“清理”（Clean Up）功能——这是苹果照片应用中一款由 AI 驱动的对象移除工具，类似于 Google Photos 中的“魔术橡皮擦”。当时，苹果软件主管 Craig Federighi 表示，公司必须“提供准确的信息，而非幻想”，这一点至关重要。当时苹果似乎对提供更广泛的 AI 编辑工具持谨慎态度，而谷歌和三星则大步向前，推出了允许用户通过描述即可在照片中添加几乎任何内容的编辑套件——包括爆炸、毒品用具以及其他潜在的有害内容。

Now, Apple is launching its own tools for manipulating photographs using prompt descriptions. An updated version of Image Playground, Apple’s AI app for generating and editing images, notably introduces the ability to generate images in a photorealistic style. Apple says this “offers new powerful ways for users to bring their imagination to life.” 现在，苹果正在推出自己的工具，利用提示词描述来操纵照片。苹果用于生成和编辑图像的 AI 应用 Image Playground 推出了更新版本，值得注意的是，它引入了生成照片级真实风格图像的能力。苹果表示，这“为用户将想象力变为现实提供了强大的新方式”。

Image Playground allows you to modify images by describing complex changes in natural language, or by tapping, circling, or brushing over specific objects to simply move or resize them. In Apple’s keynote demonstration, we saw Image Playground being used to generate an image of a woman holding a birthday cake, using a real photograph of the person as a reference. The manipulated image doesn’t just add the cake, it also entirely replaces the original background. Until now, Apple avoided photorealistic AI generation. Image Playground previously focused on cartoon-like styles that don’t believably deepfake real people. So why did Apple change its mind? Image Playground 允许用户通过自然语言描述复杂的更改，或者通过点击、圈选、涂抹特定对象来轻松移动或调整它们的大小，从而修改图像。在苹果的主题演讲演示中，我们看到 Image Playground 被用来生成一张女性手持生日蛋糕的图片，并以该人的真实照片作为参考。处理后的图像不仅添加了蛋糕，还完全替换了原始背景。在此之前，苹果一直避免使用照片级真实的 AI 生成技术。Image Playground 此前专注于卡通风格，无法令人信服地伪造真实人物。那么，苹果为什么改变了主意？

The answer, seemingly, is SynthID: Google’s near-invisible watermarking system that tags content generated by its own AI tools. Apple says any photos adjusted with Apple Intelligence will be embedded with SynthID to make them easier to identify as AI manipulated. Apple was already labeling the metadata of images that were edited using Clean Up or generated through Image Playground, but using its own “forensics” feature that, to my knowledge, isn’t used by any other major tech platform. 答案似乎是 SynthID：这是谷歌的一种近乎隐形的数字水印系统，用于标记其 AI 工具生成的内容。苹果表示，任何使用 Apple Intelligence 调整过的照片都将嵌入 SynthID，以便更容易地识别出它们是经过 AI 处理的。苹果此前已经对使用“清理”功能编辑或通过 Image Playground 生成的图像元数据进行了标记，但使用的是其自有的“取证”功能，据我所知，其他任何主流科技平台都没有使用该功能。

SynthID watermarks will be applied to photos that are edited using Clean Up, Extend, and Spatial Reframing — the trio of Apple Intelligence-powered tools for Apple’s Photos app. The updated Clean Up tool has been given a “major upgrade” according to Apple, allowing you to remove “distractions” with “better quality and more realistic infill, even when the scene is complex.” SynthID 水印将应用于使用“清理”、“扩展”（Extend）和“空间重构”（Spatial Reframing）编辑的照片——这是苹果照片应用中由 Apple Intelligence 驱动的三大工具。据苹果称，更新后的“清理”工具进行了“重大升级”，即使在场景复杂的情况下，也能以“更好的质量和更逼真的填充”移除“干扰元素”。

The new Extend tool lets you expand an image beyond its current dimensions, using generative AI to fill in the blank spaces — just like Adobe’s Generative Expand feature in Photoshop. You can use it to turn a portrait image into a landscape one, so long as you don’t mind the fact that the manipulated background isn’t actually real. 新的“扩展”工具允许你将图像扩展到当前尺寸之外，利用生成式 AI 填充空白区域——就像 Adobe Photoshop 中的“生成式扩展”功能一样。你可以用它将人像照片变成横向照片，只要你不介意处理后的背景并非真实存在这一事实。

Spatial Reframing lets you adjust the perspective of images like a 3D scene. You can select part of a photograph and drag it around with your finger to make it look like it was taken at a different angle. Apple says that Spatial Reframing builds on the understanding of spatial models that it developed for the Vision Pro headset and that it only generates new content where the perspective has been adjusted, “ensuring the reframed photo stays consistent with the original scene.” “空间重构”功能允许你像处理 3D 场景一样调整图像的透视角度。你可以选中照片的一部分并用手指拖动，使其看起来像是从不同角度拍摄的。苹果表示，“空间重构”建立在为 Vision Pro 头显开发的空间模型理解之上，并且它仅在调整了透视的区域生成新内容，“确保重构后的照片与原始场景保持一致”。

But consistency doesn’t mean authenticity. Any image edited using Apple’s tools will be flagged with AI watermarks, and if portions of the images are synthetically generated, is it really a true reflection of reality anymore? We’ve debated this subject at length at The Verge, and Apple itself has weighed in. When Apple Intelligence was announced in 2024, Federighi said Apple was “concerned” that AI could impact how “people view photographic content as something they can rely on as indicative of reality.” 但一致性并不意味着真实性。任何使用苹果工具编辑的图像都会被标记上 AI 水印，如果图像的部分内容是合成生成的，它还能算是对现实的真实反映吗？我们在 The Verge 已经深入讨论过这个话题，苹果自己也发表过看法。当 Apple Intelligence 在 2024 年发布时，Federighi 表示苹果“担心”AI 可能会影响“人们将摄影内容视为可信赖的现实指标”的方式。

AI labels are supposed to aid with this, by providing a way for online users to distinguish between real photographs and misleading AI manipulations. Support for SynthID is expanding across the industry, having recently been adopted by OpenAI. You can check images for SynthID data by uploading them into Gemini or Google’s AI-powered Search chatbot and asking if they carry the watermark. This is not exactly intuitive, but it gives users some control over checking the authenticity of images. Online platforms are also making efforts to automatically label content that carries SynthID data so that AI manipulated images can be quickly identified wherever they’re posted. AI 标签旨在通过为在线用户提供区分真实照片和误导性 AI 处理图像的方法来解决这一问题。SynthID 的支持正在整个行业内扩展，最近已被 OpenAI 采用。你可以通过将图像上传到 Gemini 或谷歌的 AI 搜索聊天机器人并询问它们是否带有水印来检查 SynthID 数据。这并不算直观，但它确实让用户在检查图像真实性方面拥有了一定的控制权。在线平台也在努力自动标记带有 SynthID 数据的内容，以便在任何地方发布 AI 处理过的图像时都能被快速识别。