bench: real reference images (anime/film/panorama) replace synthetic gradients

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

bench/minimax_bench.py

@@ -41,8 +41,12 @@ VIDEO_VAE = "minimax_h3_video_vae_fp16.safetensors"
 AUDIO_VAE = "minimax_h3_audio_vae_fp32.safetensors"
 FL2VA_DEFAULT = "minimax_h3_fl2va_pruned_int8_convrot.safetensors"
 REF2VA_DEFAULT = "minimax_h3_ref2va_pruned_int8_convrot.safetensors"
-BENCH_IMAGE = "bench_ref_1344x768.png"  # created by scripts/pod/make-test-images.py
+# Real images fetched by scripts/pod/fetch-bench-images.py (anime characters,
-BENCH_REF_SET = [f"bench_ref_{i:02d}.png" for i in range(1, 9)]  # multi-ref pool
+# film stills, giant panoramas) — realistic detail load for the ref encoder.
+BENCH_IMAGE = "bench_movie_01.jpg"
+BENCH_REF_SET = ["bench_anime_01.jpg", "bench_anime_02.jpg", "bench_anime_03.jpg",
+                 "bench_movie_01.jpg", "bench_movie_02.jpg", "bench_movie_03.jpg",
+                 "bench_big_01.jpg", "bench_big_02.jpg"]
 
 T2V_PROMPT = (
     "Cinematic aerial shot slowly orbiting a coastal lighthouse at golden hour, "

scripts/pod/fetch-bench-images.py

@@ -0,0 +1,83 @@
+#!/usr/bin/env python3
+"""Fetch real benchmark images into /ComfyUI/input via the Wikimedia Commons API.
+
+Categories:
+  bench_anime_01..03  anime-style character illustrations (detailed subjects)
+  bench_movie_01..03  public-domain film stills (cinematic composition)
+  bench_big_01..02    very large images (6000px+), stress ref preprocessing
+
+These replace the synthetic PIL gradients for ref2v/i2v benchmarking — real
+detail levels tax the reference encoder realistically. Re-runnable; skips
+files that already exist.
+"""
+import json
+import sys
+import urllib.parse
+import urllib.request
+from pathlib import Path
+
+OUT = Path("/ComfyUI/input")
+API = "https://commons.wikimedia.org/w/api.php"
+UA = {"User-Agent": "storyteller-bench/1.0 (model benchmarking; contact: ops@storyteller.ai)"}
+
+SETS = [
+    # (prefix, count, min_width, search query)
+    ("bench_anime", 3, 1000, 'anime style character illustration filetype:bitmap'),
+    ("bench_movie", 3, 1200, 'film still 1920s screenshot filetype:bitmap'),
+    ("bench_big", 2, 6000, 'panorama cityscape night filetype:bitmap'),
+]
+
+
+def api(params):
+    q = urllib.parse.urlencode({**params, "format": "json"})
+    req = urllib.request.Request(f"{API}?{q}", headers=UA)
+    with urllib.request.urlopen(req, timeout=30) as r:
+        return json.loads(r.read())
+
+
+def search_images(query, need, min_width):
+    data = api({
+        "action": "query", "generator": "search",
+        "gsrsearch": query, "gsrnamespace": 6, "gsrlimit": 40,
+        "prop": "imageinfo", "iiprop": "url|size|mime",
+    })
+    pages = (data.get("query") or {}).get("pages", {})
+    hits = []
+    for p in sorted(pages.values(), key=lambda p: p.get("index", 99)):
+        ii = (p.get("imageinfo") or [{}])[0]
+        if ii.get("mime") in ("image/jpeg", "image/png") and ii.get("width", 0) >= min_width:
+            hits.append((ii["url"], ii["width"], ii["height"]))
+        if len(hits) >= need:
+            break
+    return hits
+
+
+def fetch(url, dest):
+    req = urllib.request.Request(url, headers=UA)
+    with urllib.request.urlopen(req, timeout=120) as r, open(dest, "wb") as f:
+        f.write(r.read())
+
+
+def main():
+    OUT.mkdir(parents=True, exist_ok=True)
+    manifest = {}
+    for prefix, count, min_width, query in SETS:
+        hits = search_images(query, count, min_width)
+        if len(hits) < count:
+            print(f"WARNING: only {len(hits)}/{count} results for {prefix} ({query})")
+        for i, (url, w, h) in enumerate(hits, start=1):
+            ext = ".png" if url.lower().endswith(".png") else ".jpg"
+            name = f"{prefix}_{i:02d}{ext}"
+            dest = OUT / name
+            if dest.exists():
+                print(f"skip {name} (exists)")
+            else:
+                print(f"fetch {name} <- {w}x{h} {url}")
+                fetch(url, dest)
+            manifest[name] = {"source": url, "width": w, "height": h}
+    (OUT / "bench_images_manifest.json").write_text(json.dumps(manifest, indent=1))
+    print(f"done: {len(manifest)} images, manifest written")
+
+
+if __name__ == "__main__":
+    main()