diff --git a/apps/web/src/app/publish/entry/[author]/[permlink]/_page.tsx b/apps/web/src/app/publish/entry/[author]/[permlink]/_page.tsx
index 01252d711f..18f1729ccc 100644
--- a/apps/web/src/app/publish/entry/[author]/[permlink]/_page.tsx
+++ b/apps/web/src/app/publish/entry/[author]/[permlink]/_page.tsx
@@ -4,6 +4,7 @@ import { PublishEditor, PublishModeHeader } from "@/app/publish/_components";
 import { usePublishEditor, usePublishState } from "@/app/publish/_hooks";
 import { useEntryDetector } from "@/app/submit/_hooks";
 import { Entry } from "@/entities";
+import { metaStringList } from "@/utils";
 import i18next from "i18next";
 import { useParams } from "next/navigation";
 import { useEffect, useState } from "react";
@@ -56,14 +57,18 @@ export default function Publish() {
         setEnrty(entry);
         setStep("edit");
         setTitle(entry.title);
-        setTags(Array.from(new Set(entry.json_metadata?.tags ?? [])));
+        setTags(Array.from(new Set(metaStringList(entry.json_metadata?.tags))));
         setContent(entry.body); // todo
         // A published post carries its own generated summary, so hand the body over with it:
         // a description that is that summary keeps following the body as the author edits.
         loadMetaDescription(entry.json_metadata?.description ?? "", entry.body);
-        entry?.json_metadata?.image && setSelectedThumbnail(entry?.json_metadata?.image[0]);
-        entry?.json_metadata?.image &&
-          setEntryImages(Array.from(new Set(entry.json_metadata?.image)));
+        // Read through metaStringList: a bare-string `image` indexes to its first
+        // CHARACTER here, and spreads into one entry per character below.
+        const metaImages = metaStringList(entry.json_metadata?.image);
+        if (metaImages.length > 0) {
+          setSelectedThumbnail(metaImages[0]);
+          setEntryImages(Array.from(new Set(metaImages)));
+        }
         entry?.json_metadata?.location && setLocation(entry?.json_metadata?.location);
 
         setEditorContent(entry.body);
diff --git a/apps/web/src/app/submit/_page.tsx b/apps/web/src/app/submit/_page.tsx
index be93a1d199..0a13d57517 100644
--- a/apps/web/src/app/submit/_page.tsx
+++ b/apps/web/src/app/submit/_page.tsx
@@ -43,7 +43,7 @@ import { error, Feedback } from "@/features/shared/feedback";
 import { Navbar } from "@/features/shared/navbar";
 import { Theme } from "@/features/shared/theme";
 import i18next from "i18next";
-import { extractMetaData, isCommunity } from "@/utils";
+import { extractMetaData, isCommunity, metaStringList } from "@/utils";
 import { Draft, Entry, RewardType } from "@/entities";
 import { TextareaAutocomplete } from "@/features/shared/textarea-autocomplete";
 import { useEntryPollExtractor } from "@/features/polls";
@@ -191,14 +191,15 @@ function Submit({ path, draftId, username, permlink, searchParams }: Props) {
   useEntryDetector(username, permlink, (entry) => {
     if (entry) {
       applyTitle(entry.title);
-      applyTags(Array.from(new Set(entry.json_metadata?.tags ?? [])));
+      applyTags(Array.from(new Set(metaStringList(entry.json_metadata?.tags))));
       setBody(entry.body);
       // A description that is the post's own summary follows the body while it is rewritten,
       // the way the composer treats one: left empty here, the publish path summarises the body
       // being saved. Anything the author wrote is kept. The old fallback read the body from
       // state, which still held whatever was in the editor before this post loaded.
       setDescription(descriptionToEdit(entry.json_metadata?.description, entry.body));
-      entry?.json_metadata?.image && setSelectedThumbnail(entry?.json_metadata?.image[0]);
+      const [firstImage] = metaStringList(entry.json_metadata?.image);
+      firstImage && setSelectedThumbnail(firstImage);
       setEditingEntry(entry);
     } else if (editingEntry) {
       setEditingEntry(null);
diff --git a/apps/web/src/features/entry-management/entry-metadata-manager/entry-metadata-builder.ts b/apps/web/src/features/entry-management/entry-metadata-manager/entry-metadata-builder.ts
index d24d3f054e..0ed04ca601 100644
--- a/apps/web/src/features/entry-management/entry-metadata-manager/entry-metadata-builder.ts
+++ b/apps/web/src/features/entry-management/entry-metadata-manager/entry-metadata-builder.ts
@@ -2,7 +2,7 @@ import { postBodySummary, proxifyImageSrc } from "@ecency/render-helper";
 import { PollSnapshot } from "../../polls";
 import appPackage from "../../../../package.json";
 import { getDimensionsFromDataUrl } from "./get-dimensions-from-data-url";
-import { extractMetaData, makeApp } from "@/utils/posting";
+import { extractMetaData, makeApp, metaStringList } from "@/utils/posting";
 import { makeEntryPath } from "@/utils/make-path";
 import { AiToolsMeta, Entry, MetaData } from "@/entities";
 import { DECENTMEMES_METADATA_VERSION } from "@/api/decentmemes";
@@ -89,13 +89,22 @@ export class EntryMetadataBuilder {
     selectedThumbnail: string | undefined,
     images?: string[]
   ): Promise<this> {
-    const { image } = this.temporaryMetadata;
+    const { image, thumbnails } = this.temporaryMetadata;
 
-    let nextImages = [...(images ?? []), ...(image ?? [])];
+    // `extend(entry)` copies json_metadata verbatim, so `image` is whatever the
+    // publishing client wrote. Spreading it directly is how a legacy post whose
+    // `image` is one bare URL string gets published back as one entry PER
+    // CHARACTER, with an image_ratios probe for each, and how a numeric or object
+    // value throws "is not iterable" and makes the edit unsaveable.
+    let nextImages = [...(images ?? []), ...metaStringList(image)];
 
     if (selectedThumbnail) {
       nextImages.unshift(selectedThumbnail);
       this.withField("thumbnails", [selectedThumbnail]);
+    } else if (thumbnails !== undefined) {
+      // Nothing new to set, so the copied value is what would be published. Write
+      // the list it was meant to be rather than the shape it arrived in.
+      this.withField("thumbnails", metaStringList(thumbnails));
     }
 
     nextImages = Array.from(new Set(nextImages)).splice(0, 9);
diff --git a/apps/web/src/specs/features/entry-metadata-legacy-shapes.spec.ts b/apps/web/src/specs/features/entry-metadata-legacy-shapes.spec.ts
new file mode 100644
index 0000000000..b1fa067074
--- /dev/null
+++ b/apps/web/src/specs/features/entry-metadata-legacy-shapes.spec.ts
@@ -0,0 +1,82 @@
+import { describe, expect, it, vi } from "vitest";
+
+// The real one probes each URL over the network, which never resolves under jsdom.
+vi.mock("@/features/entry-management/entry-metadata-manager/get-dimensions-from-data-url", () => ({
+  getDimensionsFromDataUrl: vi.fn(async (url: string) => `ratio:${url}`)
+}));
+
+import { EntryMetadataBuilder } from "@/features/entry-management/entry-metadata-manager/entry-metadata-builder";
+import type { Entry } from "@/entities";
+
+const COVER = "https://i.ecency.com/DQmX/cover.png";
+
+const legacyEntry = (json_metadata: unknown) => ({ json_metadata }) as unknown as Entry;
+
+/**
+ * The edit save path is `.extend(entry)...withSelectedThumbnail(selected)`, which copies
+ * json_metadata verbatim and then spreads its `image`. json_metadata is whatever the
+ * publishing client wrote, and a 2022 Liketu post in production carries bare strings in
+ * fields declared as lists (Sentry ECENCY-NEXT-1GQM came from one of them).
+ */
+describe("EntryMetadataBuilder on legacy metadata shapes", () => {
+  it("does not publish one image entry per character of a bare-string image", async () => {
+    const builder = await new EntryMetadataBuilder()
+      .extend(legacyEntry({ image: COVER }))
+      .withSelectedThumbnail(COVER);
+
+    expect(builder.build().image).toEqual([COVER]);
+  });
+
+  it("keeps a bare-string image when no thumbnail was chosen", async () => {
+    const builder = await new EntryMetadataBuilder()
+      .extend(legacyEntry({ image: COVER }))
+      .withSelectedThumbnail(undefined);
+
+    expect(builder.build().image).toEqual([COVER]);
+  });
+
+  it("stays saveable when image is a shape that cannot be spread", async () => {
+    // `[...(image ?? [])]` threw "is not iterable" here, so the edit could not be saved.
+    for (const image of [7, { 0: COVER }, null]) {
+      const builder = await new EntryMetadataBuilder()
+        .extend(legacyEntry({ image }))
+        .withSelectedThumbnail(COVER);
+
+      expect(builder.build().image).toEqual([COVER]);
+    }
+  });
+
+  it("drops junk entries rather than publishing them back", async () => {
+    const builder = await new EntryMetadataBuilder()
+      .extend(legacyEntry({ image: [COVER, null, 7, ""] }))
+      .withSelectedThumbnail(undefined);
+
+    expect(builder.build().image).toEqual([COVER]);
+  });
+
+  it("rewrites a malformed thumbnails field it is not replacing", async () => {
+    const builder = await new EntryMetadataBuilder()
+      .extend(legacyEntry({ image: [COVER], thumbnails: COVER }))
+      .withSelectedThumbnail(undefined);
+
+    expect(builder.build().thumbnails).toEqual([COVER]);
+  });
+
+  it("computes one image_ratios entry per real image, never per character", async () => {
+    const builder = await new EntryMetadataBuilder()
+      .extend(legacyEntry({ image: COVER }))
+      .withSelectedThumbnail(COVER);
+
+    expect(builder.build().image_ratios).toHaveLength(1);
+  });
+
+  it("leaves well-formed metadata exactly as it was", async () => {
+    const images = [COVER, "https://i.ecency.com/DQmY/second.png"];
+    const builder = await new EntryMetadataBuilder()
+      .extend(legacyEntry({ image: images, thumbnails: [COVER] }))
+      .withSelectedThumbnail(COVER);
+
+    expect(builder.build().image).toEqual(images);
+    expect(builder.build().thumbnails).toEqual([COVER]);
+  });
+});
diff --git a/apps/web/src/specs/utils/posting.spec.ts b/apps/web/src/specs/utils/posting.spec.ts
index bb1979bf46..7cc850fcd7 100644
--- a/apps/web/src/specs/utils/posting.spec.ts
+++ b/apps/web/src/specs/utils/posting.spec.ts
@@ -7,7 +7,8 @@ import {
   extractMetaData,
   makeCommentOptions,
   makeJsonMetaData,
-  makeJsonMetaDataReply
+  makeJsonMetaDataReply,
+  metaStringList
 } from "../../utils/posting";
 
 describe("Posting", () => {
@@ -205,6 +206,69 @@ describe("Posting", () => {
     expect(extractMetaData(`Source: ${url} for details`).image).toEqual([url]);
   });
 
+  /**
+   * json_metadata is whatever the publishing client wrote, so the initial metadata
+   * handed to extractMetaData is untrusted. Sentry ECENCY-NEXT-1GQM was one of these
+   * reaching `.filter` on a 2022 post whose publisher writes "" for its list fields.
+   */
+  const legacyMeta = (value: unknown) => value as Parameters<typeof extractMetaData>[1];
+
+  it("(21) extractMetadata survives a list field that is not a list", () => {
+    const url = "https://i.ecency.com/DQmX/body.png";
+    expect(() => extractMetaData(`![](${url})`, legacyMeta({ image: "" }))).not.toThrow();
+    expect(extractMetaData(`![](${url})`, legacyMeta({ image: "" })).image).toEqual([url]);
+    expect(extractMetaData(`![](${url})`, legacyMeta({ thumbnails: "" })).thumbnails).toEqual([url]);
+  });
+
+  it("(22) extractMetadata keeps a legacy image stored as a bare string", () => {
+    const stored = "https://i.ecency.com/DQmX/cover.png";
+    const body = "https://i.ecency.com/DQmY/body.png";
+
+    // Nothing in the body to recover it from: dropping it here is what strips the
+    // post's cover image the first time it is edited.
+    expect(extractMetaData("no images at all", legacyMeta({ image: stored })).image).toEqual([
+      stored
+    ]);
+    expect(extractMetaData(`![](${body})`, legacyMeta({ image: stored })).image).toEqual([
+      stored,
+      body
+    ]);
+    expect(
+      extractMetaData(`![](${body})`, legacyMeta({ thumbnails: stored })).thumbnails
+    ).toContain(stored);
+  });
+
+  it("(23) extractMetadata adds no image keys to a body that has none", () => {
+    // EntryMetadataBuilder.extractFromBody spreads this result over the metadata of
+    // every new post, so an `image: []` invented here would be published on posts
+    // that have no image at all.
+    expect(extractMetaData("plain text, no images")).toEqual({});
+    expect(extractMetaData("plain text, no images", { tags: ["x"] })).toEqual({ tags: ["x"] });
+    // A well-formed field with nothing to add is left exactly as it was.
+    const kept = ["https://i.ecency.com/DQmX/kept.png"];
+    expect(extractMetaData("no images at all", { image: kept }).image).toEqual(kept);
+  });
+
+  it("(24) extractMetadata ignores non-string entries inside a list field", () => {
+    const url = "https://i.ecency.com/DQmX/cover.png";
+    expect(
+      extractMetaData("no images at all", legacyMeta({ image: [url, null, 7, ""] })).image
+    ).toEqual([url]);
+  });
+
+  it("(25) metaStringList reads any shape a publisher may have written", () => {
+    const url = "https://i.ecency.com/DQmX/cover.png";
+    expect(metaStringList([url])).toEqual([url]);
+    expect(metaStringList(url)).toEqual([url]);
+    expect(metaStringList([url, null, 7, ""])).toEqual([url]);
+    expect(metaStringList("")).toEqual([]);
+    expect(metaStringList(undefined)).toEqual([]);
+    expect(metaStringList({ 0: url })).toEqual([]);
+    // The editors index this list and spread it. On a bare string both read characters.
+    expect(metaStringList(url)[0]).toBe(url);
+    expect(Array.from(new Set(metaStringList(url)))).toEqual([url]);
+  });
+
   it("makeJsonMetaData", () => {
     const meta = {
       image: ["http://www.xx.com/a.png", "https://img.esteem.ws/h74zrad2fh.jpg"]
diff --git a/apps/web/src/utils/posting.ts b/apps/web/src/utils/posting.ts
index f5eaaa2398..6e9b0be80b 100644
--- a/apps/web/src/utils/posting.ts
+++ b/apps/web/src/utils/posting.ts
@@ -141,6 +141,22 @@ const collectImages = (body: string, pattern: RegExp, needsExtension: boolean):
   return found;
 };
 
+/**
+ * A list field of json_metadata, read as the list it is declared to be.
+ *
+ * json_metadata is whatever the publishing client wrote, so `image`, `thumbnails`
+ * and `tags` all arrive as bare strings in the wild, and as arrays holding nulls
+ * or numbers. Reading one directly costs more than a crash: `image[0]` on a
+ * string is its first CHARACTER and `new Set(image)` is its characters, both of
+ * which pass every truthiness check and get published back to the chain.
+ */
+export const metaStringList = (value: unknown): string[] => {
+  if (Array.isArray(value)) {
+    return value.filter((item): item is string => typeof item === "string" && item.length > 0);
+  }
+  return typeof value === "string" && value.length > 0 ? [value] : [];
+};
+
 export const extractMetaData = (body: string, initialMeta: MetaData = {}): MetaData => {
   // Match images with common file extensions (including RAW formats like .arw)
   const imgReg = /https?:\/\/[^\s"']+\.(?:tiff?|jpe?g|gif|png|svg|ico|heic|webp|arw)/gi;
@@ -177,8 +193,8 @@ export const extractMetaData = (body: string, initialMeta: MetaData = {}): MetaD
       (other) => other.startsWith(url) && (other[url.length] === "?" || other[url.length] === "#")
     );
   const isStale = (url: string) => isBrokenTwin(url) || isCutShortCopy(url);
-  const existingImages = (initialMeta.image ?? []).filter((url) => !isStale(url));
-  const existingThumbnails = (initialMeta.thumbnails ?? []).filter((url) => !isStale(url));
+  const existingImages = metaStringList(initialMeta.image).filter((url) => !isStale(url));
+  const existingThumbnails = metaStringList(initialMeta.thumbnails).filter((url) => !isStale(url));
 
   const allImages = Array.from(new Set([...existingImages, ...bodyImages]));