pygpt-net 2.6.65__py3-none-any.whl → 2.6.66__py3-none-any.whl

pygpt_net/provider/api/openai/video.py

@@ -0,0 +1,562 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# ================================================== #
+# This file is a part of PYGPT package               #
+# Website: https://pygpt.net                         #
+# GitHub:  https://github.com/szczyglis-dev/py-gpt   #
+# MIT License                                        #
+# Created By  : Marcin Szczygliński                  #
+# Updated Date: 2025.12.25 20:00:00                  #
+# ================================================== #
+
+import datetime
+import io
+import json
+import mimetypes
+import os
+import time
+from typing import Optional, Dict, Any, List, Tuple
+
+from openai import OpenAI
+
+from PySide6.QtCore import QObject, Signal, QRunnable, Slot
+
+from pygpt_net.core.events import KernelEvent
+from pygpt_net.core.bridge.context import BridgeContext
+from pygpt_net.item.ctx import CtxItem
+from pygpt_net.utils import trans
+
+
+class Video:
+    """
+    OpenAI Sora 2 video generation wrapper.
+    """
+
+    MODE_GENERATE = "generate"
+    MODE_IMAGE_TO_VIDEO = "image2video"
+
+    def __init__(self, window=None):
+        self.window = window
+        self.worker: Optional[VideoWorker] = None
+
+    def generate(
+            self,
+            context: BridgeContext,
+            extra: Optional[Dict[str, Any]] = None,
+            sync: bool = True,
+    ) -> bool:
+        """
+        Generate video(s) using OpenAI Sora 2 API.
+
+        :param context: BridgeContext with prompt, model, attachments
+        :param extra: extra parameters (num, inline, aspect_ratio, duration, resolution)
+        :param sync: run synchronously (blocking) if True
+        :return: True if started
+        """
+        extra = extra or {}
+        ctx = context.ctx or CtxItem()
+        model = context.model
+        prompt = context.prompt
+        num = int(extra.get("num", 1))
+        inline = bool(extra.get("inline", False))
+
+        # decide sub-mode based on attachments (image-to-video when image is attached)
+        sub_mode = self.MODE_GENERATE
+        attachments = context.attachments or {}
+        if self._has_image_attachment(attachments):
+            sub_mode = self.MODE_IMAGE_TO_VIDEO
+
+        # model used to improve the prompt (not video model)
+        prompt_model = self.window.core.models.from_defaults()
+        tmp = self.window.core.config.get('video.prompt_model')
+        if self.window.core.models.has(tmp):
+            prompt_model = self.window.core.models.get(tmp)
+
+        worker = VideoWorker()
+        worker.window = self.window
+        worker.client = self.window.core.api.openai.get_client()  # configured client
+        worker.ctx = ctx
+        worker.mode = sub_mode
+        worker.attachments = attachments
+        worker.model = (model.id if model and getattr(model, "id", None) else "sora-2")
+        worker.input_prompt = prompt or ""
+        worker.model_prompt = prompt_model  # LLM for prompt rewriting
+        worker.system_prompt = self.window.core.prompt.get('video')
+        worker.raw = self.window.core.config.get('img_raw')
+        worker.num = num
+        worker.inline = inline
+
+        # optional params (app-level options)
+        worker.aspect_ratio = str(extra.get("aspect_ratio") or self.window.core.config.get('video.aspect_ratio') or "16:9")
+        worker.duration_seconds = int(extra.get("duration") or self.window.core.config.get('video.duration') or 8)
+        worker.resolution = (extra.get("resolution") or self.window.core.config.get('video.resolution') or "720p")
+
+        # Sora limits; one output per job
+        worker.max_per_job = 1
+
+        self.worker = worker
+        self.worker.signals.finished.connect(self.window.core.video.handle_finished)
+        self.worker.signals.finished_inline.connect(self.window.core.video.handle_finished_inline)
+        self.worker.signals.status.connect(self.window.core.video.handle_status)
+        self.worker.signals.error.connect(self.window.core.video.handle_error)
+
+        if sync or not self.window.controller.kernel.async_allowed(ctx):
+            self.worker.run()
+            return True
+
+        self.window.dispatch(KernelEvent(KernelEvent.STATE_BUSY, {"id": "video"}))
+        self.window.threadpool.start(self.worker)
+        return True
+
+    def _has_image_attachment(self, attachments: Dict[str, Any]) -> bool:
+        """Check if at least one image attachment is present."""
+        for _, att in (attachments or {}).items():
+            try:
+                p = getattr(att, "path", None)
+                if p and os.path.exists(p):
+                    mt, _ = mimetypes.guess_type(p)
+                    if mt and mt.startswith("image/"):
+                        return True
+            except Exception:
+                continue
+        return False
+
+
+class VideoSignals(QObject):
+    finished = Signal(object, list, str)         # ctx, paths, prompt
+    finished_inline = Signal(object, list, str)  # ctx, paths, prompt
+    status = Signal(object)                      # message
+    error = Signal(object)                       # exception
+
+
+class VideoWorker(QRunnable):
+    """
+    Worker that encapsulates Sora 2 async job lifecycle:
+    - POST /videos to create job
+    - Poll GET /videos/{id} until status=completed or failed
+    - GET /videos/{id}/content to download MP4 bytes
+    """
+
+    # Allowed MIME types for input_reference per API
+    ALLOWED_MIME = {"image/jpeg", "image/png", "image/webp"}
+
+    def __init__(self, *args, **kwargs):
+        super().__init__()
+        self.signals = VideoSignals()
+        self.window = None
+        self.client: Optional[OpenAI] = None
+        self.ctx: Optional[CtxItem] = None
+
+        # params
+        self.mode = Video.MODE_GENERATE
+        self.attachments: Dict[str, Any] = {}
+        self.model = "sora-2"
+        self.model_prompt = None
+        self.input_prompt = ""
+        self.system_prompt = ""
+        self.inline = False
+        self.raw = False
+        self.num = 1
+
+        # video generation params (mapped to Sora API)
+        self.aspect_ratio = "16:9"
+        self.duration_seconds = 8
+        self.resolution: str = "720p"
+
+        # Sora limits
+        self.max_per_job = 1
+
+    @Slot()
+    def run(self):
+        try:
+            # Optional prompt enhancement via app default LLM
+            if not self.raw and not self.inline and self.input_prompt:
+                try:
+                    self.signals.status.emit(trans('vid.status.prompt.wait'))
+                    bridge_context = BridgeContext(
+                        prompt=self.input_prompt,
+                        system_prompt=self.system_prompt,
+                        model=self.model_prompt,
+                        max_tokens=200,
+                        temperature=1.0,
+                    )
+                    ev = KernelEvent(KernelEvent.CALL, {'context': bridge_context, 'extra': {}})
+                    self.window.dispatch(ev)
+                    resp = ev.data.get('response')
+                    if resp:
+                        self.input_prompt = resp
+                except Exception as e:
+                    self.signals.error.emit(e)
+                    self.signals.status.emit(trans('vid.status.prompt.error') + ": " + str(e))
+
+            # Sora API accepts a single video per create call; honor app's num but cap to 1 per job
+            _ = max(1, min(self.num, self.max_per_job))
+
+            # Build request parameters
+            seconds = self._clamp_seconds(self.duration_seconds)
+            size = self._resolve_size(self.aspect_ratio, self.resolution, self.model)
+
+            # Image-to-video: first image attachment as input_reference, must match "size"
+            image_path = self._first_image_attachment(self.attachments) if self.mode == Video.MODE_IMAGE_TO_VIDEO else None
+
+            self.signals.status.emit(trans('vid.status.generating') + f": {self.input_prompt}...")
+
+            # Create job
+            create_kwargs: Dict[str, Any] = {
+                "model": self.model or "sora-2",
+                "prompt": self.input_prompt or "",
+                "seconds": str(seconds),
+                "size": size,
+            }
+
+            # Attach image as input_reference; auto-resize to match requested size if needed
+            file_handle = None
+            if image_path:
+                prepared = self._prepare_input_reference(image_path, size)
+                if prepared is not None:
+                    # tuple (filename, bytes, mime) supported by OpenAI Python SDK
+                    create_kwargs["input_reference"] = prepared
+                else:
+                    # Fallback: use original file (may fail if size/mime mismatch)
+                    try:
+                        file_handle = open(image_path, "rb")
+                        create_kwargs["input_reference"] = file_handle
+                    except Exception as e:
+                        self.signals.error.emit(e)
+
+            job = self.client.videos.create(**create_kwargs)
+            video_id = self._safe_get(job, "id")
+            if not video_id:
+                # include raw payload for debugging
+                raise RuntimeError("Video job ID missing in create response.")
+
+            # Poll until completed (or failed/canceled)
+            last_progress = None
+            last_status = None
+            while True:
+                time.sleep(5)
+                job = self.client.videos.retrieve(video_id)
+                status = self._safe_get(job, "status") or ""
+                progress = self._safe_get(job, "progress")
+
+                if status != last_status or (progress is not None and progress != last_progress):
+                    try:
+                        pr_txt = f" [{int(progress)}%]" if isinstance(progress, (int, float)) else ""
+                        self.signals.status.emit(f"{trans('vid.status.generating')} {status}{pr_txt}")
+                    except Exception:
+                        pass
+                    last_progress = progress
+                    last_status = status
+
+                if status in ("completed", "failed", "canceled"):
+                    break
+
+            if file_handle is not None:
+                try:
+                    file_handle.close()
+                except Exception:
+                    pass
+
+            if status != "completed":
+                # Extract detailed reason, surface policy hints if present
+                reason = self._format_job_error(job)
+                if reason:
+                    self.signals.status.emit(reason)
+                raise RuntimeError(f"Video generation did not complete: {status}. {reason or ''}".strip())
+
+            # Download content
+            response = self.client.videos.download_content(video_id=video_id)
+            data = response.read()  # bytes stream per OpenAI Python SDK
+
+            # Save file to user video dir
+            paths: List[str] = []
+            p = self._save(0, data)
+            if p:
+                paths.append(p)
+
+            if self.inline:
+                self.signals.finished_inline.emit(self.ctx, paths, self.input_prompt)
+            else:
+                self.signals.finished.emit(self.ctx, paths, self.input_prompt)
+
+        except Exception as e:
+            self.signals.error.emit(e)
+        finally:
+            self._cleanup()
+
+    # ---------- helpers ----------
+
+    @staticmethod
+    def _clamp_seconds(requested: int) -> int:
+        """
+        Clamp duration to Sora allowed values (4, 8, 12).
+        """
+        allowed = [4, 8, 12]
+        try:
+            r = int(requested or 8)
+        except Exception:
+            r = 8
+        return min(allowed, key=lambda x: abs(x - r))
+
+    @staticmethod
+    def _parse_size(size: str) -> Tuple[int, int]:
+        """
+        Parse "WxH" string into integers; returns (W, H). Raises ValueError on invalid format.
+        """
+        parts = str(size or "").lower().split("x")
+        if len(parts) != 2:
+            raise ValueError(f"Invalid size format: {size}")
+        w = int(parts[0].strip())
+        h = int(parts[1].strip())
+        if w <= 0 or h <= 0:
+            raise ValueError(f"Invalid size values: {size}")
+        return w, h
+
+    @staticmethod
+    def _resolve_size(aspect_ratio: str, resolution: str, model_id: str) -> str:
+        """
+        Map app-level aspect ratio and resolution to Sora size enumerations.
+        Supported sizes:
+          - sora-2:       1280x720, 720x1280
+          - sora-2-pro:   1280x720, 720x1280, 1024x1792, 1792x1024
+
+        Fallback to 1280x720 or 720x1280 depending on orientation.
+        """
+        ar = (aspect_ratio or "16:9").strip().lower()
+        res = (resolution or "720p").lower().strip()
+        model = (model_id or "").lower()
+        portrait = ar in ("9:16", "9x16", "portrait")
+
+        if "sora-2-pro" in model:
+            if "1024" in res or "1792" in res or "hd" in res:
+                return "1024x1792" if portrait else "1792x1024"
+
+        return "720x1280" if portrait else "1280x720"
+
+    def _first_image_attachment(self, attachments: Dict[str, Any]) -> Optional[str]:
+        """Return path of the first image attachment, if any."""
+        for _, att in (attachments or {}).items():
+            try:
+                p = getattr(att, "path", None)
+                if p and os.path.exists(p):
+                    mt, _ = mimetypes.guess_type(p)
+                    if mt and mt.startswith("image/"):
+                        return p
+            except Exception:
+                continue
+        return None
+
+    def _prepare_input_reference(self, image_path: str, size: str) -> Optional[Tuple[str, bytes, str]]:
+        """
+        Ensure the input image matches required WxH and mime. Returns a (filename, bytes, mime) tuple.
+        - Converts on the fly using Pillow if dimensions do not match.
+        - Strips EXIF metadata and normalizes orientation.
+        - Fit mode configurable via `video.image_to_video.fit`: crop | pad | stretch (default crop).
+        """
+        try:
+            w, h = self._parse_size(size)
+        except Exception:
+            self.signals.status.emit(f"Invalid target size: {size}, using default 1280x720")
+            w, h = 1280, 720
+
+        # Try Pillow import lazily
+        try:
+            from PIL import Image, ImageOps
+        except Exception:
+            # Fallback: pass original file; may still fail if size mismatches.
+            try:
+                with open(image_path, "rb") as f:
+                    data = f.read()
+                mime = (mimetypes.guess_type(image_path)[0] or "image/jpeg")
+                if mime not in self.ALLOWED_MIME:
+                    mime = "image/jpeg"
+                filename = os.path.basename(image_path)
+                return filename, data, mime
+            except Exception:
+                return None
+
+        try:
+            img = Image.open(image_path)
+            img = ImageOps.exif_transpose(img)  # honor EXIF orientation
+
+            # If already exact size and mime allowed -> pass-through with re-encode to strip EXIF
+            if img.width == w and img.height == h:
+                buf = io.BytesIO()
+                # choose output mime based on original extension but restrict to allowed set
+                target_mime = (mimetypes.guess_type(image_path)[0] or "image/jpeg")
+                if target_mime not in self.ALLOWED_MIME:
+                    target_mime = "image/jpeg"
+
+                fmt = "WEBP" if target_mime == "image/webp" else "PNG" if target_mime == "image/png" else "JPEG"
+                if fmt == "JPEG" and img.mode not in ("RGB", "L"):
+                    img = img.convert("RGB")
+                save_kwargs = {"optimize": True}
+                if fmt == "JPEG":
+                    save_kwargs.update({"quality": 95})
+                img.save(buf, format=fmt, **save_kwargs)
+                data = buf.getvalue()
+                buf.close()
+                filename = os.path.basename(image_path) if fmt != "JPEG" else "input.jpg"
+                return filename, data, target_mime
+
+            fit_mode = str(self.window.core.config.get('video.image_to_video.fit') or "crop").lower()
+            if fit_mode not in ("crop", "pad", "stretch"):
+                fit_mode = "crop"
+
+            # Convert to RGB for JPEG output by default
+            if img.mode not in ("RGB", "L"):
+                img = img.convert("RGB")
+
+            if fit_mode == "stretch":
+                resized = img.resize((w, h), Image.Resampling.LANCZOS)
+            elif fit_mode == "pad":
+                scale = min(w / img.width, h / img.height)
+                new_w = max(1, int(img.width * scale))
+                new_h = max(1, int(img.height * scale))
+                tmp = img.resize((new_w, new_h), Image.Resampling.LANCZOS)
+                canvas = Image.new("RGB", (w, h), color=(0, 0, 0))
+                left = (w - new_w) // 2
+                top = (h - new_h) // 2
+                canvas.paste(tmp, (left, top))
+                resized = canvas
+            else:  # crop (fill then center-crop)
+                scale = max(w / img.width, h / img.height)
+                new_w = max(1, int(img.width * scale))
+                new_h = max(1, int(img.height * scale))
+                tmp = img.resize((new_w, new_h), Image.Resampling.LANCZOS)
+                left = max(0, (new_w - w) // 2)
+                top = max(0, (new_h - h) // 2)
+                resized = tmp.crop((left, top, left + w, top + h))
+
+            # Encode as JPEG (safe default per API)
+            buf = io.BytesIO()
+            resized.save(buf, format="JPEG", quality=95, optimize=True)
+            data = buf.getvalue()
+            buf.close()
+
+            self.signals.status.emit(f"Auto-resized input image to {w}x{h} (mode={fit_mode}).")
+            return "input.jpg", data, "image/jpeg"
+
+        except Exception:
+            # As a last resort, send original image
+            try:
+                with open(image_path, "rb") as f:
+                    data = f.read()
+                mime = (mimetypes.guess_type(image_path)[0] or "image/jpeg")
+                if mime not in self.ALLOWED_MIME:
+                    mime = "image/jpeg"
+                filename = os.path.basename(image_path)
+                self.signals.status.emit("Image preprocessing failed; sending original file (may fail if size/policy mismatch).")
+                return filename, data, mime
+            except Exception:
+                return None
+
+    def _save(self, idx: int, data: Optional[bytes]) -> Optional[str]:
+        """Save video bytes to file and return path."""
+        if not data:
+            return None
+        name = (
+            datetime.date.today().strftime("%Y-%m-%d") + "_" +
+            datetime.datetime.now().strftime("%H-%M-%S") + "-" +
+            self.window.core.video.make_safe_filename(self.input_prompt) + "-" +
+            str(idx + 1) + ".mp4"
+        )
+        path = os.path.join(self.window.core.config.get_user_dir("video"), name)
+        self.signals.status.emit(trans('vid.status.downloading') + f" ({idx + 1} / 1) -> {path}")
+
+        if self.window.core.video.save_video(path, data):
+            return str(path)
+
+        try:
+            os.makedirs(os.path.dirname(path), exist_ok=True)
+            with open(path, "wb") as f:
+                f.write(data)
+            return str(path)
+        except Exception:
+            return None
+
+    def _cleanup(self):
+        """Cleanup resources."""
+        sig = self.signals
+        self.signals = None
+        if sig is not None:
+            try:
+                sig.deleteLater()
+            except RuntimeError:
+                pass
+
+    # ---------- error utilities ----------
+
+    @staticmethod
+    def _safe_get(obj: Any, key: str, default=None):
+        """
+        Access attribute or dict key safely.
+        """
+        try:
+            if isinstance(obj, dict):
+                return obj.get(key, default)
+            if hasattr(obj, key):
+                return getattr(obj, key)
+        except Exception:
+            return default
+        return default
+
+    def _format_job_error(self, job: Any) -> str:
+        """
+        Build a human-readable error message from the video job.
+        Includes error.message/type/code and common policy hints.
+        """
+        # Extract error payload
+        err = self._safe_get(job, "error")
+        details: Dict[str, Any] = {}
+        message = ""
+        code = ""
+        etype = ""
+
+        # Normalize to dict
+        if err is not None:
+            try:
+                if isinstance(err, dict):
+                    details = err
+                elif hasattr(err, "model_dump"):
+                    details = err.model_dump()  # pydantic v2
+                elif hasattr(err, "to_dict"):
+                    details = err.to_dict()
+                else:
+                    # last resort string
+                    message = str(err)
+            except Exception:
+                message = str(err)
+
+        message = details.get("message", message) if isinstance(details, dict) else message
+        code = details.get("code", code) if isinstance(details, dict) else code
+        etype = details.get("type", etype) if isinstance(details, dict) else etype
+
+        # Policy-related hints
+        hint = ""
+        lower = (message or "").lower()
+        if any(x in lower for x in ("public figure", "likeness", "real person", "people", "face", "biometric")):
+            hint = "Policy: input images with real people or public figures are blocked. Use a non-real/persona or generated image."
+        elif any(x in lower for x in ("copyright", "trademark", "brand", "ip infringement")):
+            hint = "Policy: potential IP restriction detected. Avoid brand logos or copyrighted assets."
+        elif "inpaint" in lower and "width and height" in lower:
+            hint = "The reference image must exactly match the requested size (size=WxH). The app auto-resizes now; verify your aspect and chosen model allow this size."
+
+        parts = []
+        if message:
+            parts.append(f"reason: {message}")
+        if etype:
+            parts.append(f"type: {etype}")
+        if code:
+            parts.append(f"code: {code}")
+        if hint:
+            parts.append(hint)
+
+        # Include minimal structured fallback if nothing above was present
+        if not parts and isinstance(details, dict) and details:
+            try:
+                parts.append(json.dumps(details, ensure_ascii=False))
+            except Exception:
+                pass
+
+        return " ".join(parts).strip()

pygpt_net/provider/core/config/patch.py

@@ -176,6 +176,13 @@ class Patch:
                 patch_css('style.dark.css', True)
                 updated = True
 
+            # < 2.6.66
+            if old < parse_version("2.6.66"):
+                print("Migrating config from < 2.6.66...")
+                if "img_mode" not in data:
+                    data["img_mode"] = "image"
+                updated = True
+
         # update file
         migrated = False
         if updated:

pygpt_net/provider/core/model/patch.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.09.12 00:00:00                  #
+# Updated Date: 2025.12.25 20:00:00                  #
 # ================================================== #
 
 from packaging.version import parse as parse_version, Version
@@ -52,11 +52,37 @@ class Patch:
                 data, updated = patcher.execute(version)
             # --------------------------------------------
 
-            # > 2.6.42 below:
-            # pass
+            # <  2.6.66 <--- add models
+            if old < parse_version("2.6.66"):
+                print("Migrating models from < 2.6.66...")
+                models_to_add = [
+                    "claude-opus-4-5",
+                    "claude-sonnet-4-5",
+                    "gemini-3-flash-preview",
+                    "gemini-3-pro-image-preview",
+                    "gemini-3-pro-preview",
+                    "gpt-5.2-low",
+                    "gpt-5.2-medium",
+                    "gpt-5.2-high",
+                    "gpt-image-1.5",
+                    "nano-banana-pro-preview",
+                    "sora-2",
+                    "veo-3.1-fast-generate-preview",
+                    "veo-3.1-generate-preview"
+                ]
+                for model in models_to_add:
+                    if model not in data:
+                        base_model = from_base(model)
+                        if base_model:
+                            data[model] = base_model
+                            updated = True
 
         # update file
         if updated:
+            # fix empty/broken data
+            for key in list(data.keys()):
+                if not data[key]:
+                    del data[key]
             data = dict(sorted(data.items()))
             self.window.core.models.items = data
             self.window.core.models.save()