离线 · 无上传 · 浏览器本地完成Offline · no uploads · runs in the browser

PNG 转 Markdown——离线、一行内联、透明原样PNG to Markdown offline — inline, alpha untouched

两步：`png>html` 把文件的原始字节用 `FileReader.readAsDataURL` 编成 data URI 放进一个 `<img>`（`alt` 与 `title` 是文件名主干，去扩展名、无名时 `converted`，并做属性转义），`html>md` 再用 Turndown 序列化。实测产物是**一行图片语法**：`body` 里只有那张 `<img>`，样式表在 `<head>` 而序列化只取 `body`，CSS 不会漏进 `.md`。PNG 在这一路上真正被保住的是**透明**：这条链不重新采样、也不往画布上画，所以 alpha 通道与像素都是原样；对照 PNG→JPG 那一趟会把它合成掉。代价与边界都摆在明处：一，base64 每三个字节写成四个字符，再加图片语法本身，`.md` 会明显变重，整段负载还挤在同一行——要提交进仓库、要被 diff，外链是更好的形状；二，**显示由收件端决定**，网页端邮件客户端与一些托管站不画 `data:` 图片，本页不替它们承诺；三，类型前提也在这儿——从 ZIP 解包出来的文件不带 MIME 类型，那条 data URI 的前缀会是 `application/octet-stream`，而不是 `image/png`（这条路不受影响，但同一种字节再往 Word 走一步就会被剥掉 `src`）。想要能搜索、能复制的文字得靠识别工具：本扩展不带 OCR 模型与语言包，也不把图上传给任何服务，产物里唯一的字符串来自你的文件名。全程只有 storage 权限，不联网。Two steps: `png>html` encodes the file’s raw bytes with `FileReader.readAsDataURL` into a data URI inside one `<img>` (`alt` and `title` carry the basename of your file, extension dropped, `converted` when there is no usable name, escaped for attribute position), and `html>md` then serialises it with Turndown. Measured, the artifact is **one line of image syntax**: that `<img>` is all the body holds, the stylesheet sits in `<head>` and the serialisation reads only the body, so no CSS leaks in. What PNG actually keeps on this route is **its alpha channel** — the chain neither resamples nor redraws a canvas, so transparency and pixels travel byte for byte; the contrast is PNG to JPG, which composites that away. Costs and boundaries stated up front. First, base64 writes four characters per three bytes and the image syntax adds its own weight, so the `.md` gets noticeably heavier with the whole payload on a single line — for anything that goes into a repository and gets diffed, an external reference is the better shape. Second, **display belongs to the recipient**: webmail clients and some hosting sites will not draw a `data:` picture, and this page makes no promise for them. Third, the type caveat: a file unpacked from a ZIP carries no MIME type, so its data URI prefix is `application/octet-stream` rather than `image/png` (harmless on this route, but the same bytes one step toward Word do get their `src` stripped). Searchable, copyable words would need recognition: this extension ships no OCR model or language packs and uploads nothing anywhere, so the only string in the artifact is your file name. All of it on the storage permission, with no network at all.