Add an attributed real-photo image example

README.md

@@ -54,6 +54,48 @@ print(result["answers"]["answer"])
 
 Images are supplied as PNG, JPEG, or WebP data URLs (up to four images; 10 MiB each; 25 megapixels each). Requests without `images` use the selected text adapter. Requests with `images` use the visual sidecar; **the M3 text LoRA is not applied to image reads**.
 
+## Real-photo image example
+
+This example sends a real photograph through JEVision's image route. The photo shows a laptop beside a cup of coffee.
+
+![A cup of coffee beside a laptop on a wooden table](examples/real-photo/coffee-and-laptop.jpg)
+
+Photo credit: [Shixart1985](https://commons.wikimedia.org/wiki/User:Shixart1985), via Wikimedia Commons, [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/). [Source photograph](https://commons.wikimedia.org/wiki/File:Coffee_cup_next_to_laptop_on_wooden_table_in_cozy_indoor_workspace_during_daytime.jpg); the bundled file is the 960-pixel Commons thumbnail, unaltered here.
+
+The runnable request definition is [`input.json`](examples/real-photo/input.json). It names the image file so the example stays readable; [`run_demo.py`](examples/real-photo/run_demo.py) encodes that file as a data URL and posts the request to `/v1/systemone`:
+
+```bash
+python run_jevvision.py --text-adapter jevbench-m3 --port 8009
+python examples/real-photo/run_demo.py --endpoint http://127.0.0.1:8009
+```
+
+Captured response for this exact image and request:
+
+```json
+{
+  "model": "kev-latest",
+  "answers": {
+    "scene": {
+      "type": "choice",
+      "choice": "laptop_and_coffee",
+      "confidence": 1.0,
+      "probabilities": {
+        "laptop_and_coffee": 1.0,
+        "bicycle_and_helmet": 0.0,
+        "cat_on_sofa": 0.0
+      }
+    }
+  },
+  "usage": {
+    "input_tokens": 674,
+    "output_tokens": 67
+  },
+  "latency_ms": 14771.8
+}
+```
+
+The full response is saved as [`output.json`](examples/real-photo/output.json). This is one qualitative example, not an image benchmark or evidence of general image accuracy. The displayed latency came from a CPU smoke run and is not a performance claim.
+
 ## What is bundled
 
 | Component | Route | Notes |
@@ -68,12 +110,12 @@ The base is `Qwen/Qwen3.5-0.8B-Base` at revision `dc7cdfe2ee4154fa7e30f5b51ca41b
 ## Scope and limitations
 
 - **Context:** the configured request maximum is 80,000 tokens. This is not a 128K guarantee, and the M3 text adapter was trained on short benchmark examples rather than long-context training sequences.
-- **Images:** image requests route to a separate visual sidecar. Image-quality results are not reported here; the available functional checks used synthetic inputs and do not establish general real-photo accuracy or an advantage over stock Qwen.
+- **Images:** image requests route to a separate visual sidecar; the M3 text LoRA is not applied to image reads. The single real-photo example above is illustrative only, not a benchmark, a claim of general real-photo accuracy, or evidence of an advantage over stock Qwen.
 - **Text accuracy:** the M3 text branch was trained and selected using public JevBench data. Benchmark scores are intentionally omitted from this release description; this is not an independent evaluation or a claim of Jev parity.
 - **Latency:** this bundle is not latency-optimized; no cost or speed advantage is claimed.
 - **Architecture:** this is routed inference with separate text and visual LoRAs/heads—not one adapter that handles both text-only and image requests.
 
-`jevvision_manifest.json` records the adapter identities, hashes, routes, and configured limits. The raw training dataset, photos, benchmark scores, and Kaggle logs are not included.
+`jevvision_manifest.json` records the adapter identities, hashes, routes, configured limits, and the attributed example image. The raw training dataset, training/evaluation photos, benchmark scores, and Kaggle logs are not included.
 
 ## License
 

examples/real-photo/input.json

@@ -0,0 +1,16 @@
+{
+  "state": "Use the attached photo as visual context.",
+  "model": "kev-latest",
+  "image_file": "coffee-and-laptop.jpg",
+  "questions": {
+    "scene": {
+      "type": "choice",
+      "instructions": "Which description best matches the photo?",
+      "criteria": {
+        "laptop_and_coffee": "A laptop beside a cup of coffee on a table.",
+        "bicycle_and_helmet": "A bicycle parked beside a helmet.",
+        "cat_on_sofa": "A cat sitting on a sofa."
+      }
+    }
+  }
+}

examples/real-photo/output.json

@@ -0,0 +1,20 @@
+{
+  "model": "kev-latest",
+  "answers": {
+    "scene": {
+      "type": "choice",
+      "choice": "laptop_and_coffee",
+      "confidence": 1.0,
+      "probabilities": {
+        "laptop_and_coffee": 1.0,
+        "bicycle_and_helmet": 0.0,
+        "cat_on_sofa": 0.0
+      }
+    }
+  },
+  "usage": {
+    "input_tokens": 674,
+    "output_tokens": 67
+  },
+  "latency_ms": 14771.8
+}

examples/real-photo/photo-credit.md

@@ -0,0 +1,3 @@
+Photo: “Coffee cup next to laptop on wooden table in cozy indoor workspace during daytime” by Shixart1985, via Wikimedia Commons, licensed under [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/).
+
+Source: [Wikimedia Commons file page](https://commons.wikimedia.org/wiki/File:Coffee_cup_next_to_laptop_on_wooden_table_in_cozy_indoor_workspace_during_daytime.jpg). The bundled image is Wikimedia Commons’ 960-pixel thumbnail of the photograph; it is unaltered in this repository.

examples/real-photo/run_demo.py

@@ -0,0 +1,44 @@
+#!/usr/bin/env python3
+"""Send the bundled real-photo example to a running JEVision System One server."""
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+from urllib.request import Request, urlopen
+
+EXAMPLE_DIR = Path(__file__).resolve().parent
+REPO_ROOT = EXAMPLE_DIR.parents[1]
+sys.path.insert(0, str(REPO_ROOT))
+
+from jevvision import image_file_as_data_url  # noqa: E402
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--endpoint", default="http://127.0.0.1:8009")
+    parser.add_argument("--output", type=Path, default=EXAMPLE_DIR / "output.json")
+    args = parser.parse_args()
+
+    payload = json.loads((EXAMPLE_DIR / "input.json").read_text(encoding="utf-8"))
+    image_name = payload.pop("image_file")
+    image_path = (EXAMPLE_DIR / image_name).resolve()
+    image_path.relative_to(EXAMPLE_DIR.resolve())
+    payload["images"] = [image_file_as_data_url(image_path)]
+
+    request = Request(
+        args.endpoint.rstrip("/") + "/v1/systemone",
+        data=json.dumps(payload).encode("utf-8"),
+        headers={"content-type": "application/json"},
+        method="POST",
+    )
+    with urlopen(request, timeout=900) as response:
+        result = json.loads(response.read().decode("utf-8"))
+    rendered = json.dumps(result, indent=2, ensure_ascii=False) + "\n"
+    args.output.write_text(rendered, encoding="utf-8")
+    print(rendered, end="")
+
+
+if __name__ == "__main__":
+    main()

jevvision_manifest.json

@@ -38,7 +38,15 @@
   "visual_evidence": {
     "image_quality_results_published": false,
     "general_image_accuracy_established": false,
-    "real_photo_advantage_established": false
+    "real_photo_advantage_established": false,
+    "qualitative_example": {
+      "image": "examples/real-photo/coffee-and-laptop.jpg",
+      "input": "examples/real-photo/input.json",
+      "output": "examples/real-photo/output.json",
+      "image_sha256": "95abfa66426cb1ef5b41517545b26cf8968a2d50754f93367a827272f6855cac",
+      "license": "CC-BY-2.0",
+      "scope": "one qualitative real-photo request/response; not a benchmark"
+    }
   },
   "artifact_sha256": {
     "D:/Dark/jevvision_release_bundle_final_20260925/text/jevbench-m3/adapter_model.safetensors": "0f71c01d01cc9a82648417f3b036e329e0fe27dd44cd6847fa141b85518e1022",
@@ -54,7 +62,7 @@
   "omitted": [
     "base-model weights",
     "dataset files",
-    "photos",
+    "training/evaluation photos (except the separately attributed example above)",
     "training-state checkpoints",
     "unpromoted 4B experiments"
   ]