| from __future__ import annotations |
|
|
| import base64 |
| import json |
| import os |
| import shutil |
| import subprocess |
| import tempfile |
| from pathlib import Path |
| from typing import Any |
|
|
| import requests |
|
|
|
|
| DEFAULT_OPENROUTER_MODEL = "google/gemini-3.1-pro-preview" |
| DEFAULT_OPENROUTER_URL = "https://openrouter.ai/api/v1/chat/completions" |
| DEFAULT_OPENROUTER_SITE_URL = "http://localhost" |
| DEFAULT_OPENROUTER_APP_NAME = "low-high-new" |
|
|
|
|
| def extract_frames(video_path: str, *, frame_count: int = 4, width: int = 640) -> list[Path]: |
| tmpdir = Path(tempfile.mkdtemp(prefix="openrouter_frames_")) |
| out_pattern = tmpdir / "frame_%02d.jpg" |
| ffmpeg = shutil.which("ffmpeg") |
| if not ffmpeg: |
| try: |
| import imageio_ffmpeg |
|
|
| ffmpeg = imageio_ffmpeg.get_ffmpeg_exe() |
| except Exception as exc: |
| raise FileNotFoundError( |
| "ffmpeg was not found in PATH, and imageio-ffmpeg is not available. " |
| "Install ffmpeg or `pip install imageio-ffmpeg`." |
| ) from exc |
| cmd = [ |
| ffmpeg, |
| "-y", |
| "-i", |
| video_path, |
| "-vf", |
| f"fps=1,scale={width}:-1", |
| "-frames:v", |
| str(frame_count), |
| str(out_pattern), |
| ] |
| subprocess.run(cmd, check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL) |
| return sorted(tmpdir.glob("frame_*.jpg")) |
|
|
|
|
| def _image_to_data_url(path: Path) -> str: |
| payload = base64.b64encode(path.read_bytes()).decode("ascii") |
| return f"data:image/jpeg;base64,{payload}" |
|
|
|
|
| def _extract_json_object(text: str) -> dict[str, Any]: |
| stripped = text.strip() |
| if stripped.startswith("```"): |
| lines = stripped.splitlines() |
| if lines and lines[0].startswith("```"): |
| lines = lines[1:] |
| if lines and lines[-1].startswith("```"): |
| lines = lines[:-1] |
| stripped = "\n".join(lines).strip() |
| start = stripped.find("{") |
| end = stripped.rfind("}") |
| if start == -1 or end == -1 or end <= start: |
| raise ValueError("No JSON object found in model output.") |
| return json.loads(stripped[start : end + 1]) |
|
|
|
|
| class OpenRouterVisionModel: |
| def __init__( |
| self, |
| *, |
| model: str = DEFAULT_OPENROUTER_MODEL, |
| api_url: str = DEFAULT_OPENROUTER_URL, |
| api_key_env: str = "OPENROUTER_API_KEY", |
| timeout: int = 300, |
| max_tokens: int = 1500, |
| site_url: str | None = None, |
| app_name: str | None = None, |
| ) -> None: |
| self.model = model |
| self.api_url = api_url |
| self.timeout = timeout |
| self.max_tokens = max_tokens |
| self.site_url = (site_url or os.environ.get("OPENROUTER_SITE_URL") or DEFAULT_OPENROUTER_SITE_URL).strip() |
| self.app_name = (app_name or os.environ.get("OPENROUTER_APP_NAME") or DEFAULT_OPENROUTER_APP_NAME).strip() |
| self.api_key = os.environ.get(api_key_env, "").strip() |
| if not self.api_key: |
| raise RuntimeError(f"Missing OpenRouter API key in env var: {api_key_env}") |
|
|
| def predict_json(self, *, system_prompt: str, user_text: str, image_paths: list[str]) -> dict[str, Any]: |
| content: list[dict[str, Any]] = [{"type": "text", "text": user_text}] |
| for image_path in image_paths: |
| content.append({"type": "image_url", "image_url": {"url": _image_to_data_url(Path(image_path))}}) |
| response = requests.post( |
| self.api_url, |
| headers={ |
| "Authorization": f"Bearer {self.api_key}", |
| "Content-Type": "application/json", |
| "HTTP-Referer": self.site_url, |
| "X-Title": self.app_name, |
| }, |
| json={ |
| "model": self.model, |
| "messages": [ |
| {"role": "system", "content": system_prompt}, |
| {"role": "user", "content": content}, |
| ], |
| "temperature": 0.2, |
| "max_tokens": self.max_tokens, |
| }, |
| timeout=self.timeout, |
| ) |
| if not response.ok: |
| body = response.text.strip() |
| try: |
| payload = response.json() |
| body = json.dumps(payload, ensure_ascii=False) |
| except Exception: |
| pass |
| raise RuntimeError( |
| f"OpenRouter request failed: status={response.status_code}, model={self.model}, body={body}" |
| ) |
| payload = response.json() |
| text = payload["choices"][0]["message"]["content"] |
| return _extract_json_object(text) |
|
|