Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -9,8 +9,6 @@ from typing import List, Tuple, Optional
|
|
| 9 |
import requests
|
| 10 |
from PIL import Image, ImageFile, UnidentifiedImageError
|
| 11 |
import gradio as gr
|
| 12 |
-
# import asyncio # Not used
|
| 13 |
-
# import threading # Not directly used for core logic, implicit in Gradio's max_threads
|
| 14 |
import time
|
| 15 |
import atexit
|
| 16 |
from requests.exceptions import RequestException, HTTPError # Import for rate limiting
|
|
@@ -19,7 +17,7 @@ from requests.exceptions import RequestException, HTTPError # Import for rate li
|
|
| 19 |
DEFAULT_KEY = os.getenv("MISTRAL_API_KEY", "")
|
| 20 |
PIXTRAL_MODEL = "pixtral-12b-2409"
|
| 21 |
VIDEO_MODEL = "voxtral-mini-latest"
|
| 22 |
-
STREAM_THRESHOLD = 20 * 1024 * 1024
|
| 23 |
FFMPEG_BIN = shutil.which("ffmpeg")
|
| 24 |
IMAGE_EXTS = (".jpg", ".jpeg", ".png", ".webp", ".gif")
|
| 25 |
VIDEO_EXTS = (".mp4", ".mov", ".webm", ".mkv", ".avi", ".flv")
|
|
@@ -49,10 +47,11 @@ except ImportError:
|
|
| 49 |
def __init__(self, message, status_code=None, *args, **kwargs):
|
| 50 |
super().__init__(message, *args)
|
| 51 |
self.status_code = status_code
|
|
|
|
|
|
|
| 52 |
@property
|
| 53 |
def message(self):
|
| 54 |
-
return self.
|
| 55 |
-
|
| 56 |
|
| 57 |
# --- Temporary File Cleanup ---
|
| 58 |
_temp_preview_files_to_delete = []
|
|
@@ -75,48 +74,50 @@ def get_client(key: Optional[str] = None):
|
|
| 75 |
Returns a Mistral client instance. If the mistralai library is not installed
|
| 76 |
or the API key is missing, a mock client is returned which raises
|
| 77 |
MistralAPIException with specific messages.
|
|
|
|
|
|
|
| 78 |
"""
|
| 79 |
api_key = (key or "").strip() or DEFAULT_KEY
|
| 80 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 81 |
if not api_key:
|
| 82 |
-
|
| 83 |
-
def __init__(self, k): self.api_key = k
|
| 84 |
-
class Chat:
|
| 85 |
-
def complete(self, **kwargs):
|
| 86 |
-
raise MistralAPIException(
|
| 87 |
-
"Mistral API key is not set. Please provide it in the UI or as MISTRAL_API_KEY environment variable.",
|
| 88 |
-
status_code=401
|
| 89 |
-
)
|
| 90 |
-
chat = Chat()
|
| 91 |
-
class Files:
|
| 92 |
-
def upload(self, **kwargs):
|
| 93 |
-
raise MistralAPIException(
|
| 94 |
-
"Mistral API key is not set. Please provide it in the UI or as MISTRAL_API_KEY environment variable.",
|
| 95 |
-
status_code=401
|
| 96 |
-
)
|
| 97 |
-
files = Files()
|
| 98 |
-
return Dummy(api_key)
|
| 99 |
|
| 100 |
if Mistral is None:
|
| 101 |
-
|
| 102 |
-
def __init__(self, k): self.api_key = k
|
| 103 |
-
class Chat:
|
| 104 |
-
def complete(self, **kwargs):
|
| 105 |
-
raise MistralAPIException(
|
| 106 |
-
"Mistral client library is not installed. Please install it with 'pip install mistralai'.",
|
| 107 |
-
status_code=500
|
| 108 |
-
)
|
| 109 |
-
chat = Chat()
|
| 110 |
-
class Files:
|
| 111 |
-
def upload(self, **kwargs):
|
| 112 |
-
raise MistralAPIException(
|
| 113 |
-
"Mistral client library is not installed. Please install it with 'pip install mistralai'.",
|
| 114 |
-
status_code=500
|
| 115 |
-
)
|
| 116 |
-
files = Files()
|
| 117 |
-
return Dummy(api_key)
|
| 118 |
|
| 119 |
-
|
|
|
|
|
|
|
|
|
|
| 120 |
|
| 121 |
def is_remote(src: str) -> bool:
|
| 122 |
return bool(src) and src.startswith(("http://", "https://"))
|
|
@@ -264,18 +265,19 @@ def _get_video_info_and_timestamps(media_path: str, sample_count: int) -> Tuple[
|
|
| 264 |
timestamps: List[float] = []
|
| 265 |
if duration > 0 and sample_count > 0:
|
| 266 |
# Ensure we don't go past the video duration, or request too many samples
|
| 267 |
-
|
|
|
|
| 268 |
if actual_sample_count > 0:
|
| 269 |
step = duration / (actual_sample_count + 1)
|
| 270 |
timestamps = [step * (i + 1) for i in range(actual_sample_count)]
|
| 271 |
|
| 272 |
# If no timestamps generated or duration is 0, fall back to fixed times
|
| 273 |
if not timestamps:
|
| 274 |
-
timestamps = [0.5, 1.0, 2.0, 3.0, 4.0][:sample_count]
|
| 275 |
|
| 276 |
return info, timestamps
|
| 277 |
|
| 278 |
-
def extract_frames_for_model_and_gallery(media_path: str, sample_count: int = 5, timeout_extract: int = 15, gallery_base_h: int =
|
| 279 |
"""
|
| 280 |
Extracts frames from a video, processes them for both model input (high-res JPEG bytes)
|
| 281 |
and gallery display (smaller JPEG temp file paths), in a single pass.
|
|
@@ -304,6 +306,7 @@ def extract_frames_for_model_and_gallery(media_path: str, sample_count: int = 5,
|
|
| 304 |
fd_raw, tmp_png_path = tempfile.mkstemp(suffix=f"_frame_{i}.png")
|
| 305 |
os.close(fd_raw)
|
| 306 |
|
|
|
|
| 307 |
cmd_extract = [
|
| 308 |
FFMPEG_BIN, "-nostdin", "-y", "-ss", str(t), "-i", media_path,
|
| 309 |
"-frames:v", "1", "-pix_fmt", "rgb24", tmp_png_path,
|
|
@@ -317,14 +320,14 @@ def extract_frames_for_model_and_gallery(media_path: str, sample_count: int = 5,
|
|
| 317 |
raw_frame_bytes = f.read()
|
| 318 |
|
| 319 |
# For model: convert to high-res JPEG bytes (model expects this)
|
| 320 |
-
jpeg_model_bytes = convert_to_jpeg_bytes(raw_frame_bytes, base_h=
|
| 321 |
if jpeg_model_bytes: # Only append if conversion was successful
|
| 322 |
frames_for_model.append(jpeg_model_bytes)
|
| 323 |
else:
|
| 324 |
print(f"Warning: Failed to convert extracted frame {i+1} to JPEG for model input.")
|
| 325 |
|
| 326 |
# For gallery: convert to smaller JPEG bytes and save as new temp file
|
| 327 |
-
jpeg_gallery_bytes = convert_to_jpeg_bytes(raw_frame_bytes, base_h=gallery_base_h) #
|
| 328 |
if jpeg_gallery_bytes: # Only create temp file if conversion was successful
|
| 329 |
temp_jpeg_path = _temp_file(jpeg_gallery_bytes, suffix=f"_gallery_{i}.jpg") # _temp_file tracks this for cleanup
|
| 330 |
if temp_jpeg_path: # Only add to gallery if temp file was successfully created
|
|
@@ -353,23 +356,28 @@ def chat_complete(client, model: str, messages, timeout: int = 120, progress=Non
|
|
| 353 |
if progress is not None:
|
| 354 |
progress(0.6 + 0.01 * attempt, desc=f"Sending request to model (attempt {attempt+1}/{max_retries})...")
|
| 355 |
|
| 356 |
-
#
|
| 357 |
-
if
|
| 358 |
res = client.chat.complete(model=model, messages=messages, stream=False, timeout_ms=timeout * 1000)
|
| 359 |
-
|
| 360 |
-
|
|
|
|
| 361 |
if not api_key:
|
| 362 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 363 |
url = "https://api.mistral.ai/v1/chat/completions"
|
| 364 |
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
|
| 365 |
r = requests.post(url, json={"model": model, "messages": messages}, headers=headers, timeout=timeout)
|
| 366 |
r.raise_for_status() # Raises HTTPError for bad responses (4xx or 5xx)
|
| 367 |
res = r.json()
|
|
|
|
| 368 |
|
| 369 |
if progress is not None:
|
| 370 |
progress(0.8, desc="Model responded, parsing...")
|
| 371 |
|
| 372 |
-
choices = getattr(res, "choices", None) or (res.get("choices") if isinstance(res, dict) else [])
|
| 373 |
if not choices:
|
| 374 |
return f"Empty response from model: {res}"
|
| 375 |
|
|
@@ -413,7 +421,8 @@ def upload_file_to_mistral(client, path: str, filename: str | None = None, purpo
|
|
| 413 |
if progress is not None:
|
| 414 |
progress(0.5 + 0.01 * attempt, desc=f"Uploading file to model service (attempt {attempt+1}/{max_retries})...")
|
| 415 |
|
| 416 |
-
if
|
|
|
|
| 417 |
with open(path, "rb") as fh:
|
| 418 |
res = client.files.upload(file={"file_name": fname, "content": fh}, purpose=purpose)
|
| 419 |
fid = getattr(res, "id", None) or (res.get("id") if isinstance(res, dict) else None)
|
|
@@ -424,10 +433,12 @@ def upload_file_to_mistral(client, path: str, filename: str | None = None, purpo
|
|
| 424 |
if progress is not None:
|
| 425 |
progress(0.6, desc="Upload complete")
|
| 426 |
return fid
|
| 427 |
-
else: #
|
| 428 |
-
api_key = getattr(client, "api_key", "")
|
| 429 |
if not api_key:
|
| 430 |
-
raise
|
|
|
|
|
|
|
| 431 |
url = "https://api.mistral.ai/v1/files"
|
| 432 |
headers = {"Authorization": f"Bearer {api_key}"}
|
| 433 |
with open(path, "rb") as fh:
|
|
@@ -491,7 +502,7 @@ def analyze_image_structured(client, img_bytes: bytes, prompt: str, progress=Non
|
|
| 491 |
try:
|
| 492 |
if progress is not None:
|
| 493 |
progress(0.3, desc="Preparing image for analysis...")
|
| 494 |
-
jpeg = convert_to_jpeg_bytes(img_bytes, base_h=1024)
|
| 495 |
if not jpeg: # Handle case where convert_to_jpeg_bytes failed
|
| 496 |
return "Error: Could not convert image for analysis."
|
| 497 |
data_url = b64_bytes(jpeg, mime="image/jpeg")
|
|
@@ -528,9 +539,9 @@ def analyze_video_cohesive(client, video_path: str, prompt: str, progress=None)
|
|
| 528 |
]
|
| 529 |
result = chat_complete(client, VIDEO_MODEL, messages, progress=progress)
|
| 530 |
# If successful upload, still extract frames for gallery display
|
| 531 |
-
# Use the combined function for gallery frames
|
| 532 |
_, gallery_frame_paths = extract_frames_for_model_and_gallery(
|
| 533 |
-
video_path, sample_count=6, gallery_base_h=
|
| 534 |
)
|
| 535 |
return result, gallery_frame_paths
|
| 536 |
except Exception as e:
|
|
@@ -538,8 +549,9 @@ def analyze_video_cohesive(client, video_path: str, prompt: str, progress=None)
|
|
| 538 |
progress(0.35, desc=f"Upload failed for video ({e}). Extracting frames as fallback...")
|
| 539 |
|
| 540 |
# Use the combined extraction function for both model input and gallery display
|
|
|
|
| 541 |
frames_for_model_bytes, gallery_frame_paths = extract_frames_for_model_and_gallery(
|
| 542 |
-
video_path, sample_count=6, gallery_base_h=
|
| 543 |
)
|
| 544 |
|
| 545 |
if not frames_for_model_bytes:
|
|
@@ -576,9 +588,7 @@ def _convert_video_for_preview_if_needed(path: str) -> str:
|
|
| 576 |
return path # Cannot convert, return original
|
| 577 |
|
| 578 |
# Check if the video is already likely browser-compatible (MP4 with H.264/H.265)
|
| 579 |
-
|
| 580 |
-
# For a preview, extension check is usually sufficient.
|
| 581 |
-
if path.lower().endswith((".mp4", ".m4v")): # .mov might still need conversion
|
| 582 |
info = _ffprobe_streams(path)
|
| 583 |
if info:
|
| 584 |
video_streams = [s for s in info.get("streams", []) if s.get("codec_type") == "video"]
|
|
@@ -761,7 +771,6 @@ def create_demo():
|
|
| 761 |
return img_update_clear, video_update_clear, gallery_update_clear, \
|
| 762 |
gr.update(value="", visible=True), main_path_clear, screenshot_paths_clear, raw_media_path_clear
|
| 763 |
|
| 764 |
-
raw_bytes_for_analysis = b""
|
| 765 |
temp_raw_path_for_analysis = ""
|
| 766 |
try:
|
| 767 |
# 1. Fetch raw bytes once for both preview and analysis
|
|
@@ -789,6 +798,7 @@ def create_demo():
|
|
| 789 |
|
| 790 |
if not local_playable_path:
|
| 791 |
# If preview failed, cleanup the temp_raw_path_for_analysis as well
|
|
|
|
| 792 |
if temp_raw_path_for_analysis in _temp_preview_files_to_delete:
|
| 793 |
_temp_preview_files_to_delete.remove(temp_raw_path_for_analysis)
|
| 794 |
try: os.remove(temp_raw_path_for_analysis)
|
|
@@ -811,6 +821,12 @@ def create_demo():
|
|
| 811 |
gr.update(value=[], visible=False), gr.update(value="Video preview loaded.", visible=True), \
|
| 812 |
local_playable_path, [], temp_raw_path_for_analysis # Update raw_media_path_state
|
| 813 |
else:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 814 |
return img_update_clear, video_update_clear, gallery_update_clear, \
|
| 815 |
gr.update(value="Preview load failed: unknown playable format.", visible=True), \
|
| 816 |
main_path_clear, screenshot_paths_clear, raw_media_path_clear
|
|
@@ -829,7 +845,7 @@ def create_demo():
|
|
| 829 |
url_input.change(
|
| 830 |
fn=load_main_preview_and_clear_old,
|
| 831 |
inputs=[url_input],
|
| 832 |
-
outputs=[preview_image, preview_video, screenshot_gallery, preview_status, main_preview_path_state,
|
| 833 |
)
|
| 834 |
|
| 835 |
def worker(url: str, prompt: str, key: str, current_main_preview_path: str, raw_media_path: str, progress=gr.Progress()):
|
|
@@ -861,7 +877,8 @@ def create_demo():
|
|
| 861 |
Image.open(BytesIO(raw_bytes_for_analysis)).verify()
|
| 862 |
is_actually_image_for_analysis = True
|
| 863 |
except UnidentifiedImageError:
|
| 864 |
-
|
|
|
|
| 865 |
except Exception as e:
|
| 866 |
print(f"Warning: PIL error during image verification for raw analysis media ({raw_media_path}): {e}. Falling back to video.")
|
| 867 |
is_actually_video_for_analysis = True # Other PIL error, treat as video
|
|
@@ -870,7 +887,7 @@ def create_demo():
|
|
| 870 |
|
| 871 |
# --- Video Processing Path ---
|
| 872 |
if is_actually_video_for_analysis:
|
| 873 |
-
progress(0.25, desc="Running full
|
| 874 |
result_text, generated_screenshot_paths = analyze_video_cohesive(client, raw_media_path, prompt, progress=progress)
|
| 875 |
|
| 876 |
# --- Image Processing Path ---
|
|
|
|
| 9 |
import requests
|
| 10 |
from PIL import Image, ImageFile, UnidentifiedImageError
|
| 11 |
import gradio as gr
|
|
|
|
|
|
|
| 12 |
import time
|
| 13 |
import atexit
|
| 14 |
from requests.exceptions import RequestException, HTTPError # Import for rate limiting
|
|
|
|
| 17 |
DEFAULT_KEY = os.getenv("MISTRAL_API_KEY", "")
|
| 18 |
PIXTRAL_MODEL = "pixtral-12b-2409"
|
| 19 |
VIDEO_MODEL = "voxtral-mini-latest"
|
| 20 |
+
STREAM_THRESHOLD = 20 * 1024 * 1024 # 20MB
|
| 21 |
FFMPEG_BIN = shutil.which("ffmpeg")
|
| 22 |
IMAGE_EXTS = (".jpg", ".jpeg", ".png", ".webp", ".gif")
|
| 23 |
VIDEO_EXTS = (".mp4", ".mov", ".webm", ".mkv", ".avi", ".flv")
|
|
|
|
| 47 |
def __init__(self, message, status_code=None, *args, **kwargs):
|
| 48 |
super().__init__(message, *args)
|
| 49 |
self.status_code = status_code
|
| 50 |
+
self._message = message # Store message explicitly for property
|
| 51 |
+
|
| 52 |
@property
|
| 53 |
def message(self):
|
| 54 |
+
return self._message # Return the stored message
|
|
|
|
| 55 |
|
| 56 |
# --- Temporary File Cleanup ---
|
| 57 |
_temp_preview_files_to_delete = []
|
|
|
|
| 74 |
Returns a Mistral client instance. If the mistralai library is not installed
|
| 75 |
or the API key is missing, a mock client is returned which raises
|
| 76 |
MistralAPIException with specific messages.
|
| 77 |
+
A `is_real_client` flag is added for `chat_complete`/`upload_file_to_mistral`
|
| 78 |
+
to distinguish between a real client and a dummy.
|
| 79 |
"""
|
| 80 |
api_key = (key or "").strip() or DEFAULT_KEY
|
| 81 |
|
| 82 |
+
# Helper class for dummy chat/files service
|
| 83 |
+
class DummyMistralService:
|
| 84 |
+
def __init__(self, error_message: str, status_code: Optional[int] = None):
|
| 85 |
+
self._error_message = error_message
|
| 86 |
+
self._status_code = status_code
|
| 87 |
+
|
| 88 |
+
def complete(self, **kwargs):
|
| 89 |
+
raise MistralAPIException(self._error_message, status_code=self._status_code)
|
| 90 |
+
|
| 91 |
+
def upload(self, **kwargs):
|
| 92 |
+
raise MistralAPIException(self._error_message, status_code=self._status_code)
|
| 93 |
+
|
| 94 |
+
# Dummy client wrapper
|
| 95 |
+
class DummyClient:
|
| 96 |
+
def __init__(self, k, library_missing_msg: Optional[str] = None, api_key_missing_msg: Optional[str] = None):
|
| 97 |
+
self.api_key = k
|
| 98 |
+
self.is_real_client = False
|
| 99 |
+
|
| 100 |
+
# Determine which dummy service to provide
|
| 101 |
+
if not k: # API key missing
|
| 102 |
+
error_msg = api_key_missing_msg or "Mistral API key is not set."
|
| 103 |
+
status_code = 401
|
| 104 |
+
else: # Library missing (and key is present)
|
| 105 |
+
error_msg = library_missing_msg or "Mistral client library is not installed."
|
| 106 |
+
status_code = 500
|
| 107 |
+
|
| 108 |
+
self.chat = DummyMistralService(error_msg, status_code)
|
| 109 |
+
self.files = DummyMistralService(error_msg, status_code)
|
| 110 |
+
|
| 111 |
if not api_key:
|
| 112 |
+
return DummyClient(api_key, api_key_missing_msg="Mistral API key is not set. Please provide it in the UI or as MISTRAL_API_KEY environment variable.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 113 |
|
| 114 |
if Mistral is None:
|
| 115 |
+
return DummyClient(api_key, library_missing_msg="Mistral client library is not installed. Please install it with 'pip install mistralai'.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 116 |
|
| 117 |
+
# If Mistral is installed and API key is present, return real client
|
| 118 |
+
client = Mistral(api_key=api_key)
|
| 119 |
+
setattr(client, 'is_real_client', True) # Mark as real client
|
| 120 |
+
return client
|
| 121 |
|
| 122 |
def is_remote(src: str) -> bool:
|
| 123 |
return bool(src) and src.startswith(("http://", "https://"))
|
|
|
|
| 265 |
timestamps: List[float] = []
|
| 266 |
if duration > 0 and sample_count > 0:
|
| 267 |
# Ensure we don't go past the video duration, or request too many samples
|
| 268 |
+
# Limit samples to integer seconds to avoid issues with very short videos and excessive samples
|
| 269 |
+
actual_sample_count = min(sample_count, max(1, int(duration)))
|
| 270 |
if actual_sample_count > 0:
|
| 271 |
step = duration / (actual_sample_count + 1)
|
| 272 |
timestamps = [step * (i + 1) for i in range(actual_sample_count)]
|
| 273 |
|
| 274 |
# If no timestamps generated or duration is 0, fall back to fixed times
|
| 275 |
if not timestamps:
|
| 276 |
+
timestamps = [0.5, 1.0, 2.0, 3.0, 4.0][:sample_count] # Use fixed initial seconds
|
| 277 |
|
| 278 |
return info, timestamps
|
| 279 |
|
| 280 |
+
def extract_frames_for_model_and_gallery(media_path: str, sample_count: int = 5, timeout_extract: int = 15, gallery_base_h: int = 1080, model_base_h: int = 1024, progress=None) -> Tuple[List[bytes], List[str]]:
|
| 281 |
"""
|
| 282 |
Extracts frames from a video, processes them for both model input (high-res JPEG bytes)
|
| 283 |
and gallery display (smaller JPEG temp file paths), in a single pass.
|
|
|
|
| 306 |
fd_raw, tmp_png_path = tempfile.mkstemp(suffix=f"_frame_{i}.png")
|
| 307 |
os.close(fd_raw)
|
| 308 |
|
| 309 |
+
# ffmpeg command to extract a single frame
|
| 310 |
cmd_extract = [
|
| 311 |
FFMPEG_BIN, "-nostdin", "-y", "-ss", str(t), "-i", media_path,
|
| 312 |
"-frames:v", "1", "-pix_fmt", "rgb24", tmp_png_path,
|
|
|
|
| 320 |
raw_frame_bytes = f.read()
|
| 321 |
|
| 322 |
# For model: convert to high-res JPEG bytes (model expects this)
|
| 323 |
+
jpeg_model_bytes = convert_to_jpeg_bytes(raw_frame_bytes, base_h=model_base_h) # Use model_base_h
|
| 324 |
if jpeg_model_bytes: # Only append if conversion was successful
|
| 325 |
frames_for_model.append(jpeg_model_bytes)
|
| 326 |
else:
|
| 327 |
print(f"Warning: Failed to convert extracted frame {i+1} to JPEG for model input.")
|
| 328 |
|
| 329 |
# For gallery: convert to smaller JPEG bytes and save as new temp file
|
| 330 |
+
jpeg_gallery_bytes = convert_to_jpeg_bytes(raw_frame_bytes, base_h=gallery_base_h) # Use gallery_base_h
|
| 331 |
if jpeg_gallery_bytes: # Only create temp file if conversion was successful
|
| 332 |
temp_jpeg_path = _temp_file(jpeg_gallery_bytes, suffix=f"_gallery_{i}.jpg") # _temp_file tracks this for cleanup
|
| 333 |
if temp_jpeg_path: # Only add to gallery if temp file was successfully created
|
|
|
|
| 356 |
if progress is not None:
|
| 357 |
progress(0.6 + 0.01 * attempt, desc=f"Sending request to model (attempt {attempt+1}/{max_retries})...")
|
| 358 |
|
| 359 |
+
# Check if it's the real Mistral client or a dummy/fallback to direct requests
|
| 360 |
+
if getattr(client, 'is_real_client', False):
|
| 361 |
res = client.chat.complete(model=model, messages=messages, stream=False, timeout_ms=timeout * 1000)
|
| 362 |
+
choices = getattr(res, "choices", None)
|
| 363 |
+
else: # Fallback to direct requests if Mistral library not installed or API key missing
|
| 364 |
+
api_key = getattr(client, "api_key", "") # Get API key from the client object (real or dummy)
|
| 365 |
if not api_key:
|
| 366 |
+
# This case should ideally be caught by get_client returning a DummyClient
|
| 367 |
+
# but provides a safety net if somehow we bypass that.
|
| 368 |
+
raise MistralAPIException(
|
| 369 |
+
"Mistral API key is not set for direct HTTP request.", status_code=401
|
| 370 |
+
)
|
| 371 |
url = "https://api.mistral.ai/v1/chat/completions"
|
| 372 |
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
|
| 373 |
r = requests.post(url, json={"model": model, "messages": messages}, headers=headers, timeout=timeout)
|
| 374 |
r.raise_for_status() # Raises HTTPError for bad responses (4xx or 5xx)
|
| 375 |
res = r.json()
|
| 376 |
+
choices = res.get("choices") # For dictionary response from direct requests
|
| 377 |
|
| 378 |
if progress is not None:
|
| 379 |
progress(0.8, desc="Model responded, parsing...")
|
| 380 |
|
|
|
|
| 381 |
if not choices:
|
| 382 |
return f"Empty response from model: {res}"
|
| 383 |
|
|
|
|
| 421 |
if progress is not None:
|
| 422 |
progress(0.5 + 0.01 * attempt, desc=f"Uploading file to model service (attempt {attempt+1}/{max_retries})...")
|
| 423 |
|
| 424 |
+
# Check if it's the real Mistral client or a dummy/fallback to direct requests
|
| 425 |
+
if getattr(client, 'is_real_client', False):
|
| 426 |
with open(path, "rb") as fh:
|
| 427 |
res = client.files.upload(file={"file_name": fname, "content": fh}, purpose=purpose)
|
| 428 |
fid = getattr(res, "id", None) or (res.get("id") if isinstance(res, dict) else None)
|
|
|
|
| 433 |
if progress is not None:
|
| 434 |
progress(0.6, desc="Upload complete")
|
| 435 |
return fid
|
| 436 |
+
else: # Fallback to direct requests if Mistral library not installed or API key missing
|
| 437 |
+
api_key = getattr(client, "api_key", "")
|
| 438 |
if not api_key:
|
| 439 |
+
raise MistralAPIException(
|
| 440 |
+
"Mistral API key is not set for file upload.", status_code=401
|
| 441 |
+
)
|
| 442 |
url = "https://api.mistral.ai/v1/files"
|
| 443 |
headers = {"Authorization": f"Bearer {api_key}"}
|
| 444 |
with open(path, "rb") as fh:
|
|
|
|
| 502 |
try:
|
| 503 |
if progress is not None:
|
| 504 |
progress(0.3, desc="Preparing image for analysis...")
|
| 505 |
+
jpeg = convert_to_jpeg_bytes(img_bytes, base_h=1024) # High-res for model input
|
| 506 |
if not jpeg: # Handle case where convert_to_jpeg_bytes failed
|
| 507 |
return "Error: Could not convert image for analysis."
|
| 508 |
data_url = b64_bytes(jpeg, mime="image/jpeg")
|
|
|
|
| 539 |
]
|
| 540 |
result = chat_complete(client, VIDEO_MODEL, messages, progress=progress)
|
| 541 |
# If successful upload, still extract frames for gallery display
|
| 542 |
+
# Use the combined function for gallery frames, now with higher resolution
|
| 543 |
_, gallery_frame_paths = extract_frames_for_model_and_gallery(
|
| 544 |
+
video_path, sample_count=6, gallery_base_h=1080, model_base_h=1024, progress=progress
|
| 545 |
)
|
| 546 |
return result, gallery_frame_paths
|
| 547 |
except Exception as e:
|
|
|
|
| 549 |
progress(0.35, desc=f"Upload failed for video ({e}). Extracting frames as fallback...")
|
| 550 |
|
| 551 |
# Use the combined extraction function for both model input and gallery display
|
| 552 |
+
# Higher resolution for both model frames and gallery frames
|
| 553 |
frames_for_model_bytes, gallery_frame_paths = extract_frames_for_model_and_gallery(
|
| 554 |
+
video_path, sample_count=6, gallery_base_h=1080, model_base_h=1024, progress=progress
|
| 555 |
)
|
| 556 |
|
| 557 |
if not frames_for_model_bytes:
|
|
|
|
| 588 |
return path # Cannot convert, return original
|
| 589 |
|
| 590 |
# Check if the video is already likely browser-compatible (MP4 with H.264/H.265)
|
| 591 |
+
if path.lower().endswith((".mp4", ".m4v")):
|
|
|
|
|
|
|
| 592 |
info = _ffprobe_streams(path)
|
| 593 |
if info:
|
| 594 |
video_streams = [s for s in info.get("streams", []) if s.get("codec_type") == "video"]
|
|
|
|
| 771 |
return img_update_clear, video_update_clear, gallery_update_clear, \
|
| 772 |
gr.update(value="", visible=True), main_path_clear, screenshot_paths_clear, raw_media_path_clear
|
| 773 |
|
|
|
|
| 774 |
temp_raw_path_for_analysis = ""
|
| 775 |
try:
|
| 776 |
# 1. Fetch raw bytes once for both preview and analysis
|
|
|
|
| 798 |
|
| 799 |
if not local_playable_path:
|
| 800 |
# If preview failed, cleanup the temp_raw_path_for_analysis as well
|
| 801 |
+
# (though atexit/clear_all_files_and_ui would eventually get it)
|
| 802 |
if temp_raw_path_for_analysis in _temp_preview_files_to_delete:
|
| 803 |
_temp_preview_files_to_delete.remove(temp_raw_path_for_analysis)
|
| 804 |
try: os.remove(temp_raw_path_for_analysis)
|
|
|
|
| 821 |
gr.update(value=[], visible=False), gr.update(value="Video preview loaded.", visible=True), \
|
| 822 |
local_playable_path, [], temp_raw_path_for_analysis # Update raw_media_path_state
|
| 823 |
else:
|
| 824 |
+
# If local_playable_path exists but is not image/video, clean it up
|
| 825 |
+
if local_playable_path in _temp_preview_files_to_delete:
|
| 826 |
+
_temp_preview_files_to_delete.remove(local_playable_path)
|
| 827 |
+
try: os.remove(local_playable_path)
|
| 828 |
+
except Exception as e: print(f"Error during cleanup of unplayable temp file {local_playable_path}: {e}")
|
| 829 |
+
|
| 830 |
return img_update_clear, video_update_clear, gallery_update_clear, \
|
| 831 |
gr.update(value="Preview load failed: unknown playable format.", visible=True), \
|
| 832 |
main_path_clear, screenshot_paths_clear, raw_media_path_clear
|
|
|
|
| 845 |
url_input.change(
|
| 846 |
fn=load_main_preview_and_clear_old,
|
| 847 |
inputs=[url_input],
|
| 848 |
+
outputs=[preview_image, preview_video, screenshot_gallery, preview_status, main_preview_path_state, raw_media_path_state] # Removed screenshot_paths_state from here as it's cleared anyway.
|
| 849 |
)
|
| 850 |
|
| 851 |
def worker(url: str, prompt: str, key: str, current_main_preview_path: str, raw_media_path: str, progress=gr.Progress()):
|
|
|
|
| 877 |
Image.open(BytesIO(raw_bytes_for_analysis)).verify()
|
| 878 |
is_actually_image_for_analysis = True
|
| 879 |
except UnidentifiedImageError:
|
| 880 |
+
# If it's not an identifiable image, assume it might be a video
|
| 881 |
+
is_actually_video_for_analysis = True
|
| 882 |
except Exception as e:
|
| 883 |
print(f"Warning: PIL error during image verification for raw analysis media ({raw_media_path}): {e}. Falling back to video.")
|
| 884 |
is_actually_video_for_analysis = True # Other PIL error, treat as video
|
|
|
|
| 887 |
|
| 888 |
# --- Video Processing Path ---
|
| 889 |
if is_actually_video_for_analysis:
|
| 890 |
+
progress(0.25, desc="Running full-video analysis")
|
| 891 |
result_text, generated_screenshot_paths = analyze_video_cohesive(client, raw_media_path, prompt, progress=progress)
|
| 892 |
|
| 893 |
# --- Image Processing Path ---
|