diff --git a/apps/web/src/actions/media.ts b/apps/web/src/actions/media.ts
index 69db040..ce2d2d2 100644
--- a/apps/web/src/actions/media.ts
+++ b/apps/web/src/actions/media.ts
@@ -10,7 +10,7 @@ import { db } from "@/lib/db";
 import { env } from "@/env";
 import { getSeededOperator } from "@/lib/operator";
 import { checkRateLimit } from "@/lib/rate-limit";
-import { validateForWhatsApp } from "@/lib/whatsapp-media";
+import { sniffUnsupportedImage, validateForWhatsApp } from "@/lib/whatsapp-media";
 
 async function rateLimit(key: string) {
   const h = await headers();
@@ -39,6 +39,16 @@ export async function uploadMediaAction(
 
   const op = await getSeededOperator();
   const buffer = Buffer.from(await file.arrayBuffer());
+  // Magic-byte check: catches HEIC/AVIF that iOS Safari sometimes
+  // uploads with a misleading mime like image/jpeg. Mime-only check
+  // missed those — the bot then failed to extract a thumbnail and
+  // the message went out broken on the receiving end.
+  if (sniffUnsupportedImage(buffer)) {
+    return {
+      ok: false,
+      error: "Images are not supported, please re-upload images",
+    };
+  }
   const sha256 = createHash("sha256").update(buffer).digest("hex");
   const storagePath = newMediaPath(file.name);
   const absolute = absoluteMediaPath(storagePath, env.MEDIA_DIR);
diff --git a/apps/web/src/lib/recurrence.test.ts b/apps/web/src/lib/recurrence.test.ts
index 7fc81f6..3c15152 100644
--- a/apps/web/src/lib/recurrence.test.ts
+++ b/apps/web/src/lib/recurrence.test.ts
@@ -2,6 +2,7 @@ import { describe, it, expect } from "vitest";
 import { DateTime } from "luxon";
 import {
   buildRrule,
+  describeCronRule,
   describeRecurrence,
   isoWeekdayToCron,
   kindFromRrule,
@@ -233,3 +234,92 @@ describe("describeRecurrence", () => {
     ).toBe("Every day, until 2026-06-01");
   });
 });
+
+describe("describeCronRule — humanise stored cron strings", () => {
+  it("daily", () => {
+    expect(describeCronRule("CRON:32 11 * * *")).toBe("Every day at 11:32");
+    // The CRON: prefix is optional in the helper.
+    expect(describeCronRule("0 9 * * *")).toBe("Every day at 09:00");
+  });
+
+  it("weekly — single weekday and multi-day list, in canonical order", () => {
+    // Cron weekday: 1=Mon..6=Sat, 0=Sun.
+    expect(describeCronRule("CRON:0 9 * * 1")).toBe("Every week on Mon at 09:00");
+    expect(describeCronRule("CRON:0 17 * * 5,1,3")).toBe(
+      "Every week on Mon, Wed, Fri at 17:00",
+    );
+  });
+
+  it("weekly — Sunday encodes as cron 0 → ISO 7", () => {
+    expect(describeCronRule("CRON:30 8 * * 0")).toBe("Every week on Sun at 08:30");
+  });
+
+  it("weekly — range expands (1-5 → Mon..Fri)", () => {
+    expect(describeCronRule("CRON:0 9 * * 1-5")).toBe(
+      "Every week on Mon, Tue, Wed, Thu, Fri at 09:00",
+    );
+  });
+
+  it("monthly — single day uses singular 'day'", () => {
+    expect(describeCronRule("CRON:0 9 15 * *")).toBe("Every month on day 15 at 09:00");
+  });
+
+  it("monthly — multiple days uses plural 'days'", () => {
+    expect(describeCronRule("CRON:0 9 1,15,28 * *")).toBe(
+      "Every month on days 1, 15, 28 at 09:00",
+    );
+  });
+
+  it("monthly — long lists collapse with '+N more'", () => {
+    expect(describeCronRule("CRON:0 9 1,2,3,4,5,6,7,8 * *")).toBe(
+      "Every month on days 1, 2, 3, 4, 5, 6 +2 more at 09:00",
+    );
+  });
+
+  it("yearly — single month + single day", () => {
+    expect(describeCronRule("CRON:0 9 13 5 *")).toBe(
+      "Every year in May on day 13 at 09:00",
+    );
+  });
+
+  it("yearly — multi-month + multi-day Cartesian product", () => {
+    expect(describeCronRule("CRON:30 17 1,15 1,4,7,10 *")).toBe(
+      "Every year in Jan, Apr, Jul, Oct on days 1, 15 at 17:30",
+    );
+  });
+
+  it("multi-line rules join with ' · '", () => {
+    expect(describeCronRule("CRON:0 9 * * 1\n0 17 * * 5")).toBe(
+      "Every week on Mon at 09:00 · Every week on Fri at 17:00",
+    );
+  });
+
+  it("multi-line: blank lines are ignored", () => {
+    expect(describeCronRule("CRON:0 9 * * 1\n\n0 17 * * 5\n")).toBe(
+      "Every week on Mon at 09:00 · Every week on Fri at 17:00",
+    );
+  });
+
+  it("empty / not-configured rule reads useful", () => {
+    expect(describeCronRule("CRON:")).toBe("Cron (not configured)");
+    expect(describeCronRule("")).toBe("Cron (not configured)");
+  });
+
+  it("unrecognised cron shape falls back to the raw expression (better than silent)", () => {
+    // Sub-hour fields like "*/5 * * * *" aren't shapes the picker
+    // produces, but if one ever sneaks in we surface it raw rather
+    // than dropping it on the floor.
+    expect(describeCronRule("CRON:*/5 * * * *")).toBe("*/5 * * * *");
+  });
+
+  it("describeRecurrence delegates to describeCronRule for kind=cron", () => {
+    const spec: RecurrenceSpec = {
+      kind: "cron",
+      interval: 1,
+      weeklyDays: [],
+      cron: "0 9 * * *",
+      end: { kind: "never" },
+    };
+    expect(describeRecurrence(spec, FIRST)).toBe("Every day at 09:00");
+  });
+});
diff --git a/apps/web/src/lib/recurrence.ts b/apps/web/src/lib/recurrence.ts
index 8d9933d..49b08f9 100644
--- a/apps/web/src/lib/recurrence.ts
+++ b/apps/web/src/lib/recurrence.ts
@@ -120,7 +120,7 @@ const FREQ_UNIT: Record<string, string> = {
 export function describeRecurrence(spec: RecurrenceSpec, firstFire: DateTime): string {
   if (spec.kind === "none") return "One-off";
   if (spec.kind === "cron") {
-    return spec.cron ? `Cron: ${spec.cron}` : "Cron (not configured)";
+    return spec.cron ? describeCronRule(spec.cron) : "Cron (not configured)";
   }
 
   const interval = clampInterval(spec.interval);
@@ -219,3 +219,140 @@ export function kindFromRrule(rrule: string | null | undefined): RecurrenceKind
 export function isoWeekdayToCron(iso: number): number {
   return iso === 7 ? 0 : iso;
 }
+
+// ---------------------------------------------------------------------------
+// describeCronRule
+//
+// Turns the cron expressions the recurrence picker produces back into the
+// kind of human-readable sentence the rest of the app already uses for
+// RRULE-based recurrences. Used by `describeRecurrence` for cron rules
+// and by anywhere else the app surfaces a recurrence summary (reminder
+// list rows, detail page, review step).
+//
+// Shapes the picker emits, all with `MM HH` at the front:
+//
+//   "MM HH * * *"           → daily, every day
+//   "MM HH * * D[,D...]"    → weekly, on listed cron weekdays (0=Sun..6=Sat)
+//   "MM HH d[,d...] * *"    → monthly, on listed days
+//   "MM HH d[,d...] m[,m...] *" → yearly, listed days × listed months
+//
+// Multi-line rules ("CRON:line1\nline2\n…") render as the joined
+// per-line descriptions, separated by " · " for compactness in lists.
+// ---------------------------------------------------------------------------
+
+const MONTH_NAMES_FULL = [
+  "January", "February", "March", "April", "May", "June",
+  "July", "August", "September", "October", "November", "December",
+];
+
+function pad2(n: number): string {
+  return n.toString().padStart(2, "0");
+}
+
+function formatDaysList(days: number[]): string {
+  if (days.length <= 6) return days.join(", ");
+  return `${days.slice(0, 6).join(", ")} +${days.length - 6} more`;
+}
+
+function describeWeekdayList(cronDays: number[]): string {
+  // Cron weekday: 0=Sun..6=Sat. Sort + map to short labels.
+  const labels = cronDays
+    .slice()
+    .sort((a, b) => a - b)
+    .map((c) => {
+      const iso = c === 0 ? 7 : c;
+      return WEEKDAY_LABELS[iso - 1]?.short ?? "";
+    })
+    .filter(Boolean);
+  return labels.join(", ");
+}
+
+/** Describe a single cron expression (no `CRON:` prefix). */
+function describeCronExpr(expr: string): string {
+  const head = expr.match(/^(\d+)\s+(\d+)\s+(.*)$/);
+  if (!head) return expr; // unrecognised — fall back to raw
+  const minute = Number(head[1]);
+  const hour = Number(head[2]);
+  if (
+    !Number.isFinite(minute) ||
+    !Number.isFinite(hour) ||
+    minute < 0 ||
+    minute > 59 ||
+    hour < 0 ||
+    hour > 23
+  ) {
+    return expr;
+  }
+  const time = `${pad2(hour)}:${pad2(minute)}`;
+  const rest = head[3]!.trim();
+
+  // Daily: "* * *"
+  if (rest === "* * *") return `Every day at ${time}`;
+
+  // Weekly: "* * <DOWs>"
+  let m = rest.match(/^\* \* ([0-9,\-]+)$/);
+  if (m) {
+    const days = m[1]!
+      .split(",")
+      .flatMap((p) => {
+        const r = p.match(/^(\d+)-(\d+)$/);
+        if (r) {
+          const out: number[] = [];
+          for (let i = Number(r[1]); i <= Number(r[2]); i++) out.push(i);
+          return out;
+        }
+        return [Number(p)];
+      })
+      .filter((n) => n >= 0 && n <= 6);
+    const labels = describeWeekdayList(days);
+    if (!labels) return expr;
+    return `Every week on ${labels} at ${time}`;
+  }
+
+  // Monthly: "<DOMs> * *"
+  m = rest.match(/^([0-9,]+) \* \*$/);
+  if (m) {
+    const days = m[1]!
+      .split(",")
+      .map((s) => Number(s))
+      .filter((n) => n >= 1 && n <= 31);
+    if (days.length === 0) return expr;
+    return `Every month on day${days.length > 1 ? "s" : ""} ${formatDaysList(days)} at ${time}`;
+  }
+
+  // Yearly: "<DOMs> <Months> *"
+  m = rest.match(/^([0-9,]+) ([0-9,]+) \*$/);
+  if (m) {
+    const days = m[1]!
+      .split(",")
+      .map((s) => Number(s))
+      .filter((n) => n >= 1 && n <= 31);
+    const months = m[2]!
+      .split(",")
+      .map((s) => Number(s))
+      .filter((n) => n >= 1 && n <= 12);
+    if (days.length === 0 || months.length === 0) return expr;
+    const monthLabels = months.map((mo) => MONTH_NAMES_FULL[mo - 1]?.slice(0, 3)).join(", ");
+    return `Every year in ${monthLabels} on day${days.length > 1 ? "s" : ""} ${formatDaysList(days)} at ${time}`;
+  }
+
+  // Unrecognised shape — show the raw expression so the user can still
+  // make sense of it; better than swallowing entirely.
+  return expr;
+}
+
+/**
+ * Render a cron rule as a sentence. Handles:
+ *   - the optional "CRON:" prefix our app stores
+ *   - multi-line rules (joined with " · ")
+ */
+export function describeCronRule(rule: string): string {
+  const prefix = "CRON:";
+  const stripped = rule.startsWith(prefix) ? rule.slice(prefix.length) : rule;
+  const lines = stripped
+    .split("\n")
+    .map((s) => s.trim())
+    .filter(Boolean);
+  if (lines.length === 0) return "Cron (not configured)";
+  return lines.map(describeCronExpr).join(" · ");
+}
diff --git a/apps/web/src/lib/whatsapp-media.test.ts b/apps/web/src/lib/whatsapp-media.test.ts
index ade4616..8739ed2 100644
--- a/apps/web/src/lib/whatsapp-media.test.ts
+++ b/apps/web/src/lib/whatsapp-media.test.ts
@@ -3,6 +3,7 @@ import {
   classifyMediaKind,
   formatBytes,
   isUnsupportedImageMime,
+  sniffUnsupportedImage,
   validateForWhatsApp,
   WA_LIMITS,
   WA_MAX_BYTES,
@@ -131,6 +132,65 @@ describe("isUnsupportedImageMime / HEIC/HEIF/AVIF guard", () => {
   });
 });
 
+describe("sniffUnsupportedImage / magic-byte detection", () => {
+  // Helper: build a 12-byte ISOBMFF header with the chosen brand.
+  // Bytes 0..3 are the box size (any 32-bit value), 4..7 are "ftyp",
+  // 8..11 are the 4-char major brand we want the sniffer to spot.
+  function isobmffHeader(brand: string): Uint8Array {
+    const buf = new Uint8Array(12);
+    // Size — pick any 4 bytes; the sniffer ignores them.
+    buf[0] = 0x00;
+    buf[1] = 0x00;
+    buf[2] = 0x00;
+    buf[3] = 0x18;
+    // 'ftyp'
+    buf[4] = 0x66;
+    buf[5] = 0x74;
+    buf[6] = 0x79;
+    buf[7] = 0x70;
+    // Brand
+    for (let i = 0; i < 4 && i < brand.length; i++) {
+      buf[8 + i] = brand.charCodeAt(i);
+    }
+    return buf;
+  }
+
+  it("flags every HEIF/AVIF brand the bot's Sharp can't decode", () => {
+    for (const brand of ["heic", "heix", "hevc", "heim", "heis", "mif1", "msf1", "avif", "avis"]) {
+      expect(sniffUnsupportedImage(isobmffHeader(brand))).toBe(true);
+    }
+  });
+
+  it("brand match is case-insensitive (some encoders write upper-case)", () => {
+    expect(sniffUnsupportedImage(isobmffHeader("HEIC"))).toBe(true);
+    expect(sniffUnsupportedImage(isobmffHeader("Avif"))).toBe(true);
+  });
+
+  it("does NOT flag a JPEG (no 'ftyp' marker at offset 4)", () => {
+    // JPEG starts with FF D8 FF E0 ... no 'ftyp' anywhere near byte 4.
+    const jpeg = new Uint8Array([0xff, 0xd8, 0xff, 0xe0, 0x00, 0x10, 0x4a, 0x46, 0x49, 0x46, 0x00, 0x01]);
+    expect(sniffUnsupportedImage(jpeg)).toBe(false);
+  });
+
+  it("does NOT flag a PNG", () => {
+    const png = new Uint8Array([0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a, 0x00, 0x00, 0x00, 0x0d]);
+    expect(sniffUnsupportedImage(png)).toBe(false);
+  });
+
+  it("does NOT flag an unrelated ftyp brand (mp4)", () => {
+    // ISOBMFF with brand 'mp42' is a regular MP4 video container; not
+    // something Sharp would try to thumbnail in the image path. Keep
+    // it out of the unsupported set.
+    expect(sniffUnsupportedImage(isobmffHeader("mp42"))).toBe(false);
+    expect(sniffUnsupportedImage(isobmffHeader("isom"))).toBe(false);
+  });
+
+  it("returns false for tiny / truncated buffers (< 12 bytes)", () => {
+    expect(sniffUnsupportedImage(new Uint8Array(0))).toBe(false);
+    expect(sniffUnsupportedImage(new Uint8Array([0x66, 0x74, 0x79, 0x70]))).toBe(false);
+  });
+});
+
 describe("WA_MAX_BYTES is the largest single-kind cap", () => {
   it("equals the document cap (100 MB)", () => {
     expect(WA_MAX_BYTES).toBe(WA_LIMITS.document);
diff --git a/apps/web/src/lib/whatsapp-media.ts b/apps/web/src/lib/whatsapp-media.ts
index c87f406..6d79405 100644
--- a/apps/web/src/lib/whatsapp-media.ts
+++ b/apps/web/src/lib/whatsapp-media.ts
@@ -67,6 +67,50 @@ export function isUnsupportedImageMime(mimeType: string): boolean {
   return UNSUPPORTED_IMAGE_MIMES.has(mimeType.toLowerCase());
 }
 
+/**
+ * Sniff the file's magic bytes to detect HEIF / AVIF regardless of
+ * what the Content-Type header claims. iOS Safari sometimes uploads
+ * HEIC photos with mime "image/jpeg" — pure mime checks let those
+ * through, the bot's Sharp can't decode the bytes, and the message
+ * is sent without a thumbnail (or rejected by WhatsApp clients on
+ * the receiving end).
+ *
+ * Both HEIF and AVIF are ISOBMFF containers. Bytes 4..7 are the
+ * literal "ftyp" box header, followed by a 4-char brand at 8..11.
+ * The brands we care about:
+ *   HEIF: heic, heix, hevc, heim, heis, mif1, msf1
+ *   AVIF: avif, avis
+ *
+ * Returns true if the bytes look like HEIF/AVIF (and so the upload
+ * should be rejected even when the mime claimed it was JPEG/PNG).
+ */
+const UNSUPPORTED_BRANDS: ReadonlySet<string> = new Set([
+  "heic",
+  "heix",
+  "hevc",
+  "heim",
+  "heis",
+  "mif1",
+  "msf1",
+  "avif",
+  "avis",
+]);
+
+export function sniffUnsupportedImage(bytes: Uint8Array): boolean {
+  if (bytes.length < 12) return false;
+  // "ftyp" at bytes 4..7
+  if (
+    bytes[4] !== 0x66 || // 'f'
+    bytes[5] !== 0x74 || // 't'
+    bytes[6] !== 0x79 || // 'y'
+    bytes[7] !== 0x70    // 'p'
+  ) {
+    return false;
+  }
+  const brand = String.fromCharCode(bytes[8]!, bytes[9]!, bytes[10]!, bytes[11]!).toLowerCase();
+  return UNSUPPORTED_BRANDS.has(brand);
+}
+
 export type WaSizeCheck =
   | { ok: true; kind: WaMediaKind; limitBytes: number }
   | { ok: false; kind: WaMediaKind; limitBytes: number; error: string };