diff --git a/tests/test_artist_directory_service.py b/tests/test_artist_directory_service.py
index fbfbf2f..56f490e 100644
--- a/tests/test_artist_directory_service.py
+++ b/tests/test_artist_directory_service.py
@@ -5,6 +5,8 @@ ordering by (name ASC, id ASC), substring q filter, platform EXISTS
 filter (with the row-duplication trap), pagination, image_count via
 LEFT JOIN, and the window-function preview helper.
 """
+import hashlib
+
 import pytest
 
 from backend.app.models import Artist, ImageRecord, Source
@@ -58,9 +60,13 @@ async def _seed_source(db, artist_id: int, platform: str, url: str):
 
 
 async def _seed_image(db, artist_id: int, suffix: str):
+    # Derive a stable, collision-free digest from the suffix. The old
+    # `abs(hash(suffix)) % 10000` used PYTHONHASHSEED-randomized hash() over only
+    # 10k buckets, so two suffixes in one test could birthday-collide and violate
+    # uq_image_record_sha256 — flaky per process seed (CI run 1182).
     db.add(ImageRecord(
         path=f"/images/{suffix}.jpg",
-        sha256=("0" * 60) + f"{abs(hash(suffix)) % 10000:04d}",
+        sha256=hashlib.sha256(suffix.encode()).hexdigest(),
         size_bytes=10, mime="image/jpeg", width=10, height=10,
         origin="downloaded", artist_id=artist_id,
     ))