diff --git a/app/disc_images.py b/app/disc_images.py
index 7ae45f8..01baec6 100644
--- a/app/disc_images.py
+++ b/app/disc_images.py
@@ -1,24 +1,21 @@
 import hashlib
-import io
 import os
 from pathlib import Path
 
-import httpx
 from fastapi import APIRouter, HTTPException
-from fastapi.responses import FileResponse
-from PIL import Image
+from fastapi.responses import FileResponse, RedirectResponse
 from sqlalchemy import text
 
 from .db import engine
 
-# Local WebP cache for Discogs release covers. RecordGod + DealGod share a VPS, so there's no point
-# in a remote image service or a second copy — RecordGod compresses + serves covers itself.
-# Storage path is CONFIGURABLE (DISC_IMAGE_DIR env): a default now, per-store storage override later
-# when shops host their own. Covers are tiny (q72, longest side <=600px). On-demand: converted on first
-# request, then served from disk. See RECORDGOD_NAVIGATOR_PLAN / recordgod-engine.
+# Release-cover images for RecordGod. The stable URL the UI uses is /img/r/<release_id>:
+#   • if a LOCAL WebP copy exists (DISC_IMAGE_DIR), serve it — self-hosted, tiny, store-owned;
+#   • else 302 → the Discogs thumb. Discogs already serves a ~150px q40 WebP (tiny), and a BROWSER can
+#     load it even though server-side fetch is Cloudflare-403'd — so we redirect rather than proxy.
+# DISC_IMAGE_DIR is CONFIGURABLE (default now; per-store storage override later when shops host their own).
+# Local copies get populated by a backfill from WowPlatter's already-downloaded images / the harvester —
+# NOT by a server-side i.discogs fetch (that's blocked). Until then, the redirect gives working tiny WebP.
 IMAGE_DIR = Path(os.getenv("DISC_IMAGE_DIR", "/app/disc_images"))
-WEBP_QUALITY = 72
-MAX_DIM = 600
 
 router = APIRouter()
 
@@ -30,24 +27,13 @@ def _path(release_id: int) -> Path:
 
 @router.get("/img/r/{release_id}")
 async def release_image(release_id: int):
-    """Tiny WebP of a release cover — convert from disc_release.thumb on first hit, then serve from disk."""
     p = _path(release_id)
-    if not p.exists():
-        async with engine.connect() as conn:
-            row = (await conn.execute(text("SELECT thumb FROM disc_release WHERE id = :r"),
-                                      {"r": release_id})).first()
-        src = row[0] if row else None
-        if not src:
-            raise HTTPException(404, "no source image")
-        try:
-            async with httpx.AsyncClient(timeout=15, follow_redirects=True) as c:
-                r = await c.get(src, headers={"User-Agent": "Mozilla/5.0 (RecordGod)"})
-            r.raise_for_status()
-            img = Image.open(io.BytesIO(r.content)).convert("RGB")
-            img.thumbnail((MAX_DIM, MAX_DIM))
-            p.parent.mkdir(parents=True, exist_ok=True)
-            img.save(p, "WEBP", quality=WEBP_QUALITY, method=6)
-        except Exception:
-            raise HTTPException(502, "image fetch/convert failed")
-    return FileResponse(p, media_type="image/webp",
-                        headers={"Cache-Control": "public, max-age=2592000"})
+    if p.exists():
+        return FileResponse(p, media_type="image/webp",
+                            headers={"Cache-Control": "public, max-age=2592000"})
+    async with engine.connect() as conn:
+        row = (await conn.execute(text("SELECT thumb FROM disc_release WHERE id = :r"),
+                                  {"r": release_id})).first()
+    if row and row[0]:
+        return RedirectResponse(row[0], status_code=302)
+    raise HTTPException(404, "no image")
diff --git a/requirements.txt b/requirements.txt
index 1c7f24d..20a610f 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,5 +3,4 @@ uvicorn[standard]
 sqlalchemy[asyncio]
 asyncpg
 httpx
-pillow
 cryptography