Spaces:
Sleeping
Sleeping
CB commited on
Update streamlit_app.py
Browse files- streamlit_app.py +214 -193
streamlit_app.py
CHANGED
|
@@ -1,4 +1,17 @@
|
|
| 1 |
# streamlit_app.py
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
import os
|
| 3 |
import time
|
| 4 |
import string
|
|
@@ -6,7 +19,6 @@ import hashlib
|
|
| 6 |
import traceback
|
| 7 |
from glob import glob
|
| 8 |
from pathlib import Path
|
| 9 |
-
from difflib import SequenceMatcher
|
| 10 |
import json
|
| 11 |
import logging
|
| 12 |
|
|
@@ -15,19 +27,7 @@ import ffmpeg
|
|
| 15 |
import streamlit as st
|
| 16 |
from dotenv import load_dotenv
|
| 17 |
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
# Optional phi integration (Agent wrapper)
|
| 21 |
-
try:
|
| 22 |
-
from phi.agent import Agent
|
| 23 |
-
from phi.model.google import Gemini
|
| 24 |
-
from phi.tools.duckduckgo import DuckDuckGo
|
| 25 |
-
HAS_PHI = True
|
| 26 |
-
except Exception:
|
| 27 |
-
Agent = Gemini = DuckDuckGo = None
|
| 28 |
-
HAS_PHI = False
|
| 29 |
-
|
| 30 |
-
# google.generativeai SDK (try both legacy and newer patterns)
|
| 31 |
try:
|
| 32 |
import google.generativeai as genai
|
| 33 |
genai_responses = getattr(genai, "responses", None) or getattr(genai, "Responses", None)
|
|
@@ -41,13 +41,18 @@ except Exception:
|
|
| 41 |
get_file = None
|
| 42 |
HAS_GENAI = False
|
| 43 |
|
|
|
|
|
|
|
|
|
|
| 44 |
logging.basicConfig(level=logging.INFO)
|
| 45 |
logger = logging.getLogger("video_ai")
|
| 46 |
|
|
|
|
| 47 |
st.set_page_config(page_title="Generate the story of videos", layout="wide")
|
| 48 |
DATA_DIR = Path("./data")
|
| 49 |
DATA_DIR.mkdir(exist_ok=True)
|
| 50 |
|
|
|
|
| 51 |
st.session_state.setdefault("videos", "")
|
| 52 |
st.session_state.setdefault("loop_video", False)
|
| 53 |
st.session_state.setdefault("uploaded_file", None)
|
|
@@ -64,6 +69,7 @@ st.session_state.setdefault("last_url_value", "")
|
|
| 64 |
st.session_state.setdefault("processing_timeout", 900)
|
| 65 |
st.session_state.setdefault("generation_timeout", 300)
|
| 66 |
st.session_state.setdefault("preferred_model", "gemini-2.5-flash-lite")
|
|
|
|
| 67 |
|
| 68 |
MODEL_OPTIONS = [
|
| 69 |
"gemini-2.5-flash",
|
|
@@ -73,6 +79,7 @@ MODEL_OPTIONS = [
|
|
| 73 |
"custom",
|
| 74 |
]
|
| 75 |
|
|
|
|
| 76 |
def sanitize_filename(path_str: str):
|
| 77 |
name = Path(path_str).name
|
| 78 |
return name.lower().translate(str.maketrans("", "", string.punctuation)).replace(" ", "_")
|
|
@@ -95,13 +102,26 @@ def convert_video_to_mp4(video_path: str) -> str:
|
|
| 95 |
pass
|
| 96 |
return target_path
|
| 97 |
|
| 98 |
-
def compress_video(input_path: str, target_path: str, crf: int = 28, preset: str = "fast"):
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
try:
|
| 100 |
-
ffmpeg.input(input_path)
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 104 |
except Exception:
|
|
|
|
| 105 |
return input_path
|
| 106 |
|
| 107 |
def download_video_ytdlp(url: str, save_dir: str, video_password: str = None) -> str:
|
|
@@ -141,96 +161,38 @@ def configure_genai_if_needed():
|
|
| 141 |
if genai is not None and hasattr(genai, "configure"):
|
| 142 |
genai.configure(api_key=key)
|
| 143 |
except Exception:
|
| 144 |
-
|
| 145 |
return True
|
| 146 |
|
| 147 |
-
|
| 148 |
-
def
|
| 149 |
-
|
|
|
|
|
|
|
|
|
|
| 150 |
key = get_effective_api_key()
|
| 151 |
-
if not
|
| 152 |
-
|
| 153 |
-
|
| 154 |
-
|
| 155 |
-
|
| 156 |
try:
|
| 157 |
if genai is not None and hasattr(genai, "configure"):
|
| 158 |
genai.configure(api_key=key)
|
| 159 |
-
_agent = Agent(name="Video AI summarizer", model=Gemini(id=model_id), tools=[DuckDuckGo()], markdown=True)
|
| 160 |
-
st.session_state["last_model"] = model_id
|
| 161 |
except Exception:
|
| 162 |
-
|
| 163 |
-
return _agent
|
| 164 |
-
|
| 165 |
-
def clear_all_video_state():
|
| 166 |
-
st.session_state.pop("uploaded_file", None)
|
| 167 |
-
st.session_state.pop("processed_file", None)
|
| 168 |
-
st.session_state["videos"] = ""
|
| 169 |
-
st.session_state["last_loaded_path"] = ""
|
| 170 |
-
st.session_state["analysis_out"] = ""
|
| 171 |
-
st.session_state["last_error"] = ""
|
| 172 |
-
st.session_state["file_hash"] = None
|
| 173 |
-
for f in glob(str(DATA_DIR / "*")):
|
| 174 |
-
try:
|
| 175 |
-
os.remove(f)
|
| 176 |
-
except Exception:
|
| 177 |
-
pass
|
| 178 |
-
|
| 179 |
-
current_url = st.session_state.get("url", "")
|
| 180 |
-
if current_url != st.session_state.get("last_url_value"):
|
| 181 |
-
clear_all_video_state()
|
| 182 |
-
st.session_state["last_url_value"] = current_url
|
| 183 |
-
|
| 184 |
-
st.sidebar.header("Video Input")
|
| 185 |
-
st.sidebar.text_input("Video URL", key="url", placeholder="https://")
|
| 186 |
-
|
| 187 |
-
settings_exp = st.sidebar.expander("Settings", expanded=False)
|
| 188 |
-
chosen = settings_exp.selectbox("Gemini model", MODEL_OPTIONS, index=MODEL_OPTIONS.index("gemini-2.5-flash-lite"))
|
| 189 |
-
custom_model = ""
|
| 190 |
-
if chosen == "custom":
|
| 191 |
-
custom_model = settings_exp.text_input("Custom model name", value=st.session_state.get("preferred_model", "gemini-2.5-flash-lite"))
|
| 192 |
-
model_input_value = (custom_model.strip() if chosen == "custom" else chosen).strip()
|
| 193 |
-
settings_exp.text_input("Google API Key", key="api_key", value=os.getenv("GOOGLE_API_KEY", ""), type="password")
|
| 194 |
-
default_prompt = (
|
| 195 |
-
"Watch the video and provide a detailed behavioral report focusing on human actions, interactions, posture, movement, and apparent intent. Keep language professional. Include a list of observations for notable events."
|
| 196 |
-
)
|
| 197 |
-
analysis_prompt = settings_exp.text_area("Enter analysis", value=default_prompt, height=140)
|
| 198 |
-
settings_exp.text_input("Video Password (if needed)", key="video-password", placeholder="password", type="password")
|
| 199 |
-
|
| 200 |
-
settings_exp.number_input(
|
| 201 |
-
"Processing timeout (s)", min_value=60, max_value=3600,
|
| 202 |
-
value=st.session_state.get("processing_timeout", 900), step=30,
|
| 203 |
-
key="processing_timeout",
|
| 204 |
-
)
|
| 205 |
-
settings_exp.number_input(
|
| 206 |
-
"Generation timeout (s)", min_value=30, max_value=1800,
|
| 207 |
-
value=st.session_state.get("generation_timeout", 300), step=10,
|
| 208 |
-
key="generation_timeout",
|
| 209 |
-
)
|
| 210 |
-
|
| 211 |
-
key_source = "session" if st.session_state.get("api_key") else ".env" if os.getenv("GOOGLE_API_KEY") else "none"
|
| 212 |
-
settings_exp.caption(f"Using API key from: **{key_source}**")
|
| 213 |
-
if not get_effective_api_key():
|
| 214 |
-
settings_exp.warning("No Google API key provided; upload/generation disabled.", icon="⚠️")
|
| 215 |
-
|
| 216 |
-
safety_settings = [
|
| 217 |
-
{"category": "HARM_CATEGORY_HARASSMENT", "threshold": "OFF"},
|
| 218 |
-
{"category": "HARM_CATEGORY_HATE_SPEECH", "threshold": "OFF"},
|
| 219 |
-
{"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT", "threshold": "OFF"},
|
| 220 |
-
{"category": "HARM_CATEGORY_DANGEROUS_CONTENT", "threshold": "OFF"},
|
| 221 |
-
]
|
| 222 |
|
| 223 |
-
|
| 224 |
-
|
| 225 |
-
|
| 226 |
-
|
| 227 |
-
|
| 228 |
-
raise
|
| 229 |
-
if genai is not None and hasattr(genai, "configure"):
|
| 230 |
-
genai.configure(api_key=key)
|
| 231 |
-
return upload_file(filepath)
|
| 232 |
|
| 233 |
def wait_for_processed(file_obj, timeout: int = None, progress_callback=None):
|
|
|
|
|
|
|
|
|
|
| 234 |
if timeout is None:
|
| 235 |
timeout = st.session_state.get("processing_timeout", 900)
|
| 236 |
if not HAS_GENAI or get_file is None:
|
|
@@ -268,45 +230,7 @@ def wait_for_processed(file_obj, timeout: int = None, progress_callback=None):
|
|
| 268 |
time.sleep(backoff)
|
| 269 |
backoff = min(backoff * 2, 8.0)
|
| 270 |
|
| 271 |
-
|
| 272 |
-
if not prompt or not text:
|
| 273 |
-
return text
|
| 274 |
-
a = " ".join(prompt.strip().lower().split())
|
| 275 |
-
b_full = text.strip()
|
| 276 |
-
b = " ".join(b_full[:check_len].lower().split())
|
| 277 |
-
ratio = SequenceMatcher(None, a, b).ratio()
|
| 278 |
-
if ratio >= ratio_threshold:
|
| 279 |
-
cut = min(len(b_full), max(int(len(prompt) * 0.9), len(a)))
|
| 280 |
-
new_text = b_full[cut:].lstrip(" \n:-")
|
| 281 |
-
if len(new_text) >= 3:
|
| 282 |
-
return new_text
|
| 283 |
-
placeholders = ["enter analysis", "enter your analysis", "enter analysis here", "please enter analysis"]
|
| 284 |
-
low = b_full.strip().lower()
|
| 285 |
-
for ph in placeholders:
|
| 286 |
-
if low.startswith(ph):
|
| 287 |
-
return b_full[len(ph):].lstrip(" \n:-")
|
| 288 |
-
return text
|
| 289 |
-
|
| 290 |
-
def compress_video_if_large(local_path: str, threshold_mb: int = 50):
|
| 291 |
-
try:
|
| 292 |
-
file_size_mb = os.path.getsize(local_path) / (1024 * 1024)
|
| 293 |
-
except Exception as e:
|
| 294 |
-
st.session_state["last_error"] = f"Failed to stat file before compression: {e}"
|
| 295 |
-
return local_path, False
|
| 296 |
-
|
| 297 |
-
if file_size_mb <= threshold_mb:
|
| 298 |
-
return local_path, False
|
| 299 |
-
|
| 300 |
-
compressed_path = str(Path(local_path).with_name(Path(local_path).stem + "_compressed.mp4"))
|
| 301 |
-
try:
|
| 302 |
-
result = compress_video(local_path, compressed_path, crf=28, preset="fast")
|
| 303 |
-
if result and os.path.exists(result):
|
| 304 |
-
return result, True
|
| 305 |
-
return local_path, False
|
| 306 |
-
except Exception as e:
|
| 307 |
-
st.session_state["last_error"] = f"Video compression failed: {e}\n{traceback.format_exc()}"
|
| 308 |
-
return local_path, False
|
| 309 |
-
|
| 310 |
def _normalize_genai_response(response):
|
| 311 |
if response is None:
|
| 312 |
return ""
|
|
@@ -371,6 +295,7 @@ def _normalize_genai_response(response):
|
|
| 371 |
seen.add(t)
|
| 372 |
return "\n\n".join(filtered).strip()
|
| 373 |
|
|
|
|
| 374 |
def generate_via_responses_api(prompt_text: str, processed, model_used: str, max_tokens: int = 1024, timeout: int = 300, progress_callback=None):
|
| 375 |
key = get_effective_api_key()
|
| 376 |
if not key:
|
|
@@ -387,8 +312,11 @@ def generate_via_responses_api(prompt_text: str, processed, model_used: str, max
|
|
| 387 |
user_msg = {"role": "user", "content": "Please summarize the attached video."}
|
| 388 |
call_variants = []
|
| 389 |
|
| 390 |
-
|
| 391 |
-
call_variants.append({"method": "responses.
|
|
|
|
|
|
|
|
|
|
| 392 |
call_variants.append({"method": "legacy_responses_create", "payload": {"model": model_used, "input": prompt_text, "file": fname, "max_output_tokens": max_tokens}})
|
| 393 |
|
| 394 |
def is_transient_error(e_text: str):
|
|
@@ -450,7 +378,96 @@ def generate_via_responses_api(prompt_text: str, processed, model_used: str, max
|
|
| 450 |
time.sleep(backoff)
|
| 451 |
backoff = min(backoff * 2, 8.0)
|
| 452 |
|
| 453 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 454 |
col1, col2 = st.columns([1, 3])
|
| 455 |
with col1:
|
| 456 |
generate_now = st.button("Generate the story", type="primary", disabled=not bool(get_effective_api_key()))
|
|
@@ -483,7 +500,19 @@ if st.session_state["videos"]:
|
|
| 483 |
st.session_state["loop_video"] = loop_checkbox
|
| 484 |
|
| 485 |
if st.button("Clear Video(s)"):
|
| 486 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 487 |
|
| 488 |
try:
|
| 489 |
with open(st.session_state["videos"], "rb") as vf:
|
|
@@ -495,11 +524,14 @@ if st.session_state["videos"]:
|
|
| 495 |
try:
|
| 496 |
file_size_mb = os.path.getsize(st.session_state["videos"]) / (1024 * 1024)
|
| 497 |
st.sidebar.caption(f"File size: {file_size_mb:.1f} MB")
|
| 498 |
-
if file_size_mb >
|
| 499 |
st.sidebar.warning("Large file detected — it will be compressed automatically before upload.", icon="⚠️")
|
|
|
|
|
|
|
| 500 |
except Exception:
|
| 501 |
pass
|
| 502 |
|
|
|
|
| 503 |
if generate_now and not st.session_state.get("busy"):
|
| 504 |
if not st.session_state.get("videos"):
|
| 505 |
st.error("No video loaded. Use 'Load Video' in the sidebar.")
|
|
@@ -514,12 +546,12 @@ if generate_now and not st.session_state.get("busy"):
|
|
| 514 |
if HAS_GENAI and genai is not None:
|
| 515 |
genai.configure(api_key=key_to_use)
|
| 516 |
except Exception:
|
| 517 |
-
|
| 518 |
|
| 519 |
model_id = model_input_value or st.session_state.get("preferred_model") or "gemini-2.5-flash-lite"
|
| 520 |
if st.session_state.get("last_model") != model_id:
|
| 521 |
st.session_state["last_model"] = ""
|
| 522 |
-
|
| 523 |
|
| 524 |
processed = st.session_state.get("processed_file")
|
| 525 |
current_path = st.session_state.get("videos")
|
|
@@ -536,7 +568,23 @@ if generate_now and not st.session_state.get("busy"):
|
|
| 536 |
if not HAS_GENAI:
|
| 537 |
raise RuntimeError("google.generativeai SDK not available; install it.")
|
| 538 |
local_path = current_path
|
| 539 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 540 |
|
| 541 |
with st.spinner(f"Uploading video{' (compressed)' if compressed else ''}..."):
|
| 542 |
try:
|
|
@@ -574,52 +622,23 @@ if generate_now and not st.session_state.get("busy"):
|
|
| 574 |
max_tokens = 2048 if "2.5" in model_used else 1024
|
| 575 |
est_tokens = max_tokens
|
| 576 |
|
| 577 |
-
|
| 578 |
-
|
| 579 |
-
|
| 580 |
-
|
| 581 |
-
|
| 582 |
-
|
| 583 |
-
|
| 584 |
-
|
| 585 |
-
|
| 586 |
-
|
| 587 |
-
|
| 588 |
-
|
| 589 |
-
|
| 590 |
-
|
| 591 |
-
|
| 592 |
-
|
| 593 |
-
|
| 594 |
-
except Exception:
|
| 595 |
-
pass
|
| 596 |
-
if agent_text and str(agent_text).strip():
|
| 597 |
-
out = str(agent_text).strip()
|
| 598 |
-
debug_info["agent_ok"] = True
|
| 599 |
-
debug_info["agent_response_has_text"] = True
|
| 600 |
-
else:
|
| 601 |
-
debug_info["agent_ok"] = False
|
| 602 |
-
except Exception as ae:
|
| 603 |
-
debug_info["agent_error"] = f"{ae}"
|
| 604 |
-
debug_info["agent_traceback"] = traceback.format_exc()
|
| 605 |
-
|
| 606 |
-
if not out:
|
| 607 |
-
try:
|
| 608 |
-
gen_progress_placeholder = st.empty()
|
| 609 |
-
gen_status = gen_progress_placeholder.text("Starting generation...")
|
| 610 |
-
start_gen = time.time()
|
| 611 |
-
def gen_progress_cb(stage, elapsed, info):
|
| 612 |
-
try:
|
| 613 |
-
gen_status.text(f"Stage: {stage} — elapsed: {elapsed}s — {info}")
|
| 614 |
-
except Exception:
|
| 615 |
-
pass
|
| 616 |
-
out = generate_via_responses_api(prompt_text, processed, model_used, max_tokens=max_tokens, timeout=st.session_state.get("generation_timeout", 300), progress_callback=gen_progress_cb)
|
| 617 |
-
gen_progress_placeholder.text(f"Generation complete in {int(time.time()-start_gen)}s")
|
| 618 |
-
except Exception as e:
|
| 619 |
-
tb = traceback.format_exc()
|
| 620 |
-
st.session_state["last_error"] = f"Responses API error: {e}\n\nDebug: {debug_info}\n\nTraceback:\n{tb}"
|
| 621 |
-
st.error("An error occurred while generating the story. You can try Generate again; the uploaded video will be reused.")
|
| 622 |
-
out = ""
|
| 623 |
|
| 624 |
if out:
|
| 625 |
out = remove_prompt_echo(prompt_text, out)
|
|
@@ -642,17 +661,19 @@ if generate_now and not st.session_state.get("busy"):
|
|
| 642 |
|
| 643 |
except Exception as e:
|
| 644 |
tb = traceback.format_exc()
|
| 645 |
-
st.session_state["last_error"] = f"{e}\n\
|
| 646 |
st.error("An error occurred while generating the story. You can try Generate again; the uploaded video will be reused.")
|
| 647 |
finally:
|
| 648 |
st.session_state["busy"] = False
|
| 649 |
|
|
|
|
| 650 |
if st.session_state.get("analysis_out"):
|
| 651 |
just_loaded_same = (st.session_state.get("last_loaded_path") == st.session_state.get("videos"))
|
| 652 |
if not just_loaded_same:
|
| 653 |
st.subheader("Analysis Result")
|
| 654 |
st.markdown(st.session_state.get("analysis_out"))
|
| 655 |
|
|
|
|
| 656 |
if st.session_state.get("last_error"):
|
| 657 |
with st.expander("Last Error", expanded=False):
|
| 658 |
st.write(st.session_state.get("last_error"))
|
|
|
|
| 1 |
# streamlit_app.py
|
| 2 |
+
"""
|
| 3 |
+
Streamlit app for video captioning / analysis using Google GenAI Responses API.
|
| 4 |
+
Removed phi-agent support. Uses google.generativeai SDK (Responses).
|
| 5 |
+
Requires GOOGLE_API_KEY in environment or entered in UI.
|
| 6 |
+
|
| 7 |
+
Features:
|
| 8 |
+
- Download video via yt-dlp
|
| 9 |
+
- Optional compression for files > 200 MB (configurable)
|
| 10 |
+
- Upload video via google.generativeai.upload_file and wait for processing via get_file
|
| 11 |
+
- Generate analysis via Responses.generate (or Responses.create legacy compatibility)
|
| 12 |
+
- Basic UI for model selection, prompts, timeouts, and status/progress reporting
|
| 13 |
+
"""
|
| 14 |
+
|
| 15 |
import os
|
| 16 |
import time
|
| 17 |
import string
|
|
|
|
| 19 |
import traceback
|
| 20 |
from glob import glob
|
| 21 |
from pathlib import Path
|
|
|
|
| 22 |
import json
|
| 23 |
import logging
|
| 24 |
|
|
|
|
| 27 |
import streamlit as st
|
| 28 |
from dotenv import load_dotenv
|
| 29 |
|
| 30 |
+
# Google GenAI SDK
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
try:
|
| 32 |
import google.generativeai as genai
|
| 33 |
genai_responses = getattr(genai, "responses", None) or getattr(genai, "Responses", None)
|
|
|
|
| 41 |
get_file = None
|
| 42 |
HAS_GENAI = False
|
| 43 |
|
| 44 |
+
load_dotenv()
|
| 45 |
+
|
| 46 |
+
# Logging
|
| 47 |
logging.basicConfig(level=logging.INFO)
|
| 48 |
logger = logging.getLogger("video_ai")
|
| 49 |
|
| 50 |
+
# App config
|
| 51 |
st.set_page_config(page_title="Generate the story of videos", layout="wide")
|
| 52 |
DATA_DIR = Path("./data")
|
| 53 |
DATA_DIR.mkdir(exist_ok=True)
|
| 54 |
|
| 55 |
+
# Session defaults
|
| 56 |
st.session_state.setdefault("videos", "")
|
| 57 |
st.session_state.setdefault("loop_video", False)
|
| 58 |
st.session_state.setdefault("uploaded_file", None)
|
|
|
|
| 69 |
st.session_state.setdefault("processing_timeout", 900)
|
| 70 |
st.session_state.setdefault("generation_timeout", 300)
|
| 71 |
st.session_state.setdefault("preferred_model", "gemini-2.5-flash-lite")
|
| 72 |
+
st.session_state.setdefault("compression_threshold_mb", 200) # new threshold per plan
|
| 73 |
|
| 74 |
MODEL_OPTIONS = [
|
| 75 |
"gemini-2.5-flash",
|
|
|
|
| 79 |
"custom",
|
| 80 |
]
|
| 81 |
|
| 82 |
+
# Utilities
|
| 83 |
def sanitize_filename(path_str: str):
|
| 84 |
name = Path(path_str).name
|
| 85 |
return name.lower().translate(str.maketrans("", "", string.punctuation)).replace(" ", "_")
|
|
|
|
| 102 |
pass
|
| 103 |
return target_path
|
| 104 |
|
| 105 |
+
def compress_video(input_path: str, target_path: str, crf: int = 28, preset: str = "fast", bitrate: str = None):
|
| 106 |
+
"""
|
| 107 |
+
Compress video using ffmpeg; tune via crf or bitrate.
|
| 108 |
+
Returns target_path on success, else original input_path.
|
| 109 |
+
"""
|
| 110 |
try:
|
| 111 |
+
out = ffmpeg.input(input_path)
|
| 112 |
+
params = {"vcodec": "libx264", "crf": crf, "preset": preset}
|
| 113 |
+
if bitrate:
|
| 114 |
+
params["video_bitrate"] = bitrate
|
| 115 |
+
# ffmpeg-python uses keyword 'b' for bitrate if passed via output string; using bitrate via args below
|
| 116 |
+
stream = out.output(target_path, **{"vcodec": "libx264", "preset": preset}, video_bitrate=bitrate)
|
| 117 |
+
else:
|
| 118 |
+
stream = out.output(target_path, **params)
|
| 119 |
+
stream.run(overwrite_output=True, quiet=True)
|
| 120 |
+
if os.path.exists(target_path):
|
| 121 |
+
return target_path
|
| 122 |
+
return input_path
|
| 123 |
except Exception:
|
| 124 |
+
logger.exception("Compression failed")
|
| 125 |
return input_path
|
| 126 |
|
| 127 |
def download_video_ytdlp(url: str, save_dir: str, video_password: str = None) -> str:
|
|
|
|
| 161 |
if genai is not None and hasattr(genai, "configure"):
|
| 162 |
genai.configure(api_key=key)
|
| 163 |
except Exception:
|
| 164 |
+
logger.exception("Failed to configure genai")
|
| 165 |
return True
|
| 166 |
|
| 167 |
+
# Upload & processing helpers (using google.generativeai SDK functions upload_file/get_file)
|
| 168 |
+
def upload_video_sdk(filepath: str, progress_callback=None):
|
| 169 |
+
"""
|
| 170 |
+
Upload a local file using google.generativeai.upload_file.
|
| 171 |
+
Assumes genai.configure(api_key=...) was called.
|
| 172 |
+
"""
|
| 173 |
key = get_effective_api_key()
|
| 174 |
+
if not key:
|
| 175 |
+
raise RuntimeError("No API key provided")
|
| 176 |
+
if not HAS_GENAI or upload_file is None:
|
| 177 |
+
raise RuntimeError("google.generativeai SDK not available; cannot upload")
|
| 178 |
+
# SDK upload_file typically takes path and returns file object
|
| 179 |
try:
|
| 180 |
if genai is not None and hasattr(genai, "configure"):
|
| 181 |
genai.configure(api_key=key)
|
|
|
|
|
|
|
| 182 |
except Exception:
|
| 183 |
+
pass
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 184 |
|
| 185 |
+
# call upload_file and return its result
|
| 186 |
+
try:
|
| 187 |
+
return upload_file(filepath)
|
| 188 |
+
except Exception as e:
|
| 189 |
+
logger.exception("Upload failed")
|
| 190 |
+
raise
|
|
|
|
|
|
|
|
|
|
| 191 |
|
| 192 |
def wait_for_processed(file_obj, timeout: int = None, progress_callback=None):
|
| 193 |
+
"""
|
| 194 |
+
Poll get_file(name_or_id) until processing state changes away from 'PROCESSING' or timeout.
|
| 195 |
+
"""
|
| 196 |
if timeout is None:
|
| 197 |
timeout = st.session_state.get("processing_timeout", 900)
|
| 198 |
if not HAS_GENAI or get_file is None:
|
|
|
|
| 230 |
time.sleep(backoff)
|
| 231 |
backoff = min(backoff * 2, 8.0)
|
| 232 |
|
| 233 |
+
# Response normalization
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 234 |
def _normalize_genai_response(response):
|
| 235 |
if response is None:
|
| 236 |
return ""
|
|
|
|
| 295 |
seen.add(t)
|
| 296 |
return "\n\n".join(filtered).strip()
|
| 297 |
|
| 298 |
+
# Generation via Responses API (supports modern and legacy patterns)
|
| 299 |
def generate_via_responses_api(prompt_text: str, processed, model_used: str, max_tokens: int = 1024, timeout: int = 300, progress_callback=None):
|
| 300 |
key = get_effective_api_key()
|
| 301 |
if not key:
|
|
|
|
| 312 |
user_msg = {"role": "user", "content": "Please summarize the attached video."}
|
| 313 |
call_variants = []
|
| 314 |
|
| 315 |
+
# preferred modern call
|
| 316 |
+
call_variants.append({"method": "responses.generate", "payload": {"model": model_used, "messages": [system_msg, user_msg], "files": [{"name": fname}], "max_output_tokens": max_tokens}})
|
| 317 |
+
# alternate modern payload shape
|
| 318 |
+
call_variants.append({"method": "responses.generate_alt", "payload": {"model": model_used, "input": [{"text": prompt_text, "files": [{"name": fname}]}], "max_output_tokens": max_tokens}})
|
| 319 |
+
# legacy
|
| 320 |
call_variants.append({"method": "legacy_responses_create", "payload": {"model": model_used, "input": prompt_text, "file": fname, "max_output_tokens": max_tokens}})
|
| 321 |
|
| 322 |
def is_transient_error(e_text: str):
|
|
|
|
| 378 |
time.sleep(backoff)
|
| 379 |
backoff = min(backoff * 2, 8.0)
|
| 380 |
|
| 381 |
+
# Prompt echo removal
|
| 382 |
+
from difflib import SequenceMatcher
|
| 383 |
+
def remove_prompt_echo(prompt: str, text: str, check_len: int = 600, ratio_threshold: float = 0.68):
|
| 384 |
+
if not prompt or not text:
|
| 385 |
+
return text
|
| 386 |
+
a = " ".join(prompt.strip().lower().split())
|
| 387 |
+
b_full = text.strip()
|
| 388 |
+
b = " ".join(b_full[:check_len].lower().split())
|
| 389 |
+
ratio = SequenceMatcher(None, a, b).ratio()
|
| 390 |
+
if ratio >= ratio_threshold:
|
| 391 |
+
cut = min(len(b_full), max(int(len(prompt) * 0.9), len(a)))
|
| 392 |
+
new_text = b_full[cut:].lstrip(" \n:-")
|
| 393 |
+
if len(new_text) >= 3:
|
| 394 |
+
return new_text
|
| 395 |
+
placeholders = ["enter analysis", "enter your analysis", "enter analysis here", "please enter analysis"]
|
| 396 |
+
low = b_full.strip().lower()
|
| 397 |
+
for ph in placeholders:
|
| 398 |
+
if low.startswith(ph):
|
| 399 |
+
return b_full[len(ph):].lstrip(" \n:-")
|
| 400 |
+
return text
|
| 401 |
+
|
| 402 |
+
# UI
|
| 403 |
+
current_url = st.session_state.get("url", "")
|
| 404 |
+
if current_url != st.session_state.get("last_url_value"):
|
| 405 |
+
# clear per-plan
|
| 406 |
+
st.session_state["videos"] = ""
|
| 407 |
+
st.session_state["last_loaded_path"] = ""
|
| 408 |
+
st.session_state["uploaded_file"] = None
|
| 409 |
+
st.session_state["processed_file"] = None
|
| 410 |
+
st.session_state["analysis_out"] = ""
|
| 411 |
+
st.session_state["last_error"] = ""
|
| 412 |
+
st.session_state["file_hash"] = None
|
| 413 |
+
for f in glob(str(DATA_DIR / "*")):
|
| 414 |
+
try:
|
| 415 |
+
os.remove(f)
|
| 416 |
+
except Exception:
|
| 417 |
+
pass
|
| 418 |
+
st.session_state["last_url_value"] = current_url
|
| 419 |
+
|
| 420 |
+
st.sidebar.header("Video Input")
|
| 421 |
+
st.sidebar.text_input("Video URL", key="url", placeholder="https://")
|
| 422 |
+
|
| 423 |
+
settings_exp = st.sidebar.expander("Settings", expanded=False)
|
| 424 |
+
chosen = settings_exp.selectbox("Gemini model", MODEL_OPTIONS, index=MODEL_OPTIONS.index(st.session_state.get("preferred_model", "gemini-2.5-flash-lite")))
|
| 425 |
+
custom_model = ""
|
| 426 |
+
if chosen == "custom":
|
| 427 |
+
custom_model = settings_exp.text_input("Custom model name", value=st.session_state.get("preferred_model", "gemini-2.5-flash-lite"))
|
| 428 |
+
model_input_value = (custom_model.strip() if chosen == "custom" else chosen).strip()
|
| 429 |
+
|
| 430 |
+
settings_exp.text_input("Google API Key", key="api_key", value=os.getenv("GOOGLE_API_KEY", ""), type="password")
|
| 431 |
+
|
| 432 |
+
default_prompt = (
|
| 433 |
+
"Watch the video and provide a detailed behavioral report focusing on human actions, interactions, posture, movement, and apparent intent. Keep language professional. Include a list of observations for notable events."
|
| 434 |
+
)
|
| 435 |
+
analysis_prompt = settings_exp.text_area("Enter analysis prompt", value=default_prompt, height=140)
|
| 436 |
+
settings_exp.text_input("Video Password (if needed)", key="video-password", placeholder="password", type="password")
|
| 437 |
+
|
| 438 |
+
settings_exp.number_input(
|
| 439 |
+
"Processing timeout (s)", min_value=60, max_value=3600,
|
| 440 |
+
value=st.session_state.get("processing_timeout", 900), step=30,
|
| 441 |
+
key="processing_timeout",
|
| 442 |
+
)
|
| 443 |
+
settings_exp.number_input(
|
| 444 |
+
"Generation timeout (s)", min_value=30, max_value=1800,
|
| 445 |
+
value=st.session_state.get("generation_timeout", 300), step=10,
|
| 446 |
+
key="generation_timeout",
|
| 447 |
+
)
|
| 448 |
+
|
| 449 |
+
# Compression threshold control (per plan: 200 MB)
|
| 450 |
+
settings_exp.number_input(
|
| 451 |
+
"Compression threshold (MB)", min_value=10, max_value=2000,
|
| 452 |
+
value=st.session_state.get("compression_threshold_mb", 200), step=10,
|
| 453 |
+
key="compression_threshold_mb",
|
| 454 |
+
)
|
| 455 |
+
settings_exp.caption("Files ≤ threshold are uploaded unchanged. Files > threshold are compressed before upload (tunable).")
|
| 456 |
+
|
| 457 |
+
key_source = "session" if st.session_state.get("api_key") else ".env" if os.getenv("GOOGLE_API_KEY") else "none"
|
| 458 |
+
settings_exp.caption(f"Using API key from: **{key_source}**")
|
| 459 |
+
if not get_effective_api_key():
|
| 460 |
+
settings_exp.warning("No Google API key provided; upload/generation disabled.", icon="⚠️")
|
| 461 |
+
|
| 462 |
+
# Safety settings placeholder (kept minimal)
|
| 463 |
+
safety_settings = [
|
| 464 |
+
{"category": "HARM_CATEGORY_HARASSMENT", "threshold": "OFF"},
|
| 465 |
+
{"category": "HARM_CATEGORY_HATE_SPEECH", "threshold": "OFF"},
|
| 466 |
+
{"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT", "threshold": "OFF"},
|
| 467 |
+
{"category": "HARM_CATEGORY_DANGEROUS_CONTENT", "threshold": "OFF"},
|
| 468 |
+
]
|
| 469 |
+
|
| 470 |
+
# Buttons / UI layout
|
| 471 |
col1, col2 = st.columns([1, 3])
|
| 472 |
with col1:
|
| 473 |
generate_now = st.button("Generate the story", type="primary", disabled=not bool(get_effective_api_key()))
|
|
|
|
| 500 |
st.session_state["loop_video"] = loop_checkbox
|
| 501 |
|
| 502 |
if st.button("Clear Video(s)"):
|
| 503 |
+
# minimal clear
|
| 504 |
+
st.session_state["videos"] = ""
|
| 505 |
+
st.session_state["last_loaded_path"] = ""
|
| 506 |
+
st.session_state["uploaded_file"] = None
|
| 507 |
+
st.session_state["processed_file"] = None
|
| 508 |
+
st.session_state["analysis_out"] = ""
|
| 509 |
+
st.session_state["last_error"] = ""
|
| 510 |
+
st.session_state["file_hash"] = None
|
| 511 |
+
for f in glob(str(DATA_DIR / "*")):
|
| 512 |
+
try:
|
| 513 |
+
os.remove(f)
|
| 514 |
+
except Exception:
|
| 515 |
+
pass
|
| 516 |
|
| 517 |
try:
|
| 518 |
with open(st.session_state["videos"], "rb") as vf:
|
|
|
|
| 524 |
try:
|
| 525 |
file_size_mb = os.path.getsize(st.session_state["videos"]) / (1024 * 1024)
|
| 526 |
st.sidebar.caption(f"File size: {file_size_mb:.1f} MB")
|
| 527 |
+
if file_size_mb > st.session_state.get("compression_threshold_mb", 200):
|
| 528 |
st.sidebar.warning("Large file detected — it will be compressed automatically before upload.", icon="⚠️")
|
| 529 |
+
else:
|
| 530 |
+
st.sidebar.info("File ≤ threshold — will be uploaded unchanged.")
|
| 531 |
except Exception:
|
| 532 |
pass
|
| 533 |
|
| 534 |
+
# Generation flow
|
| 535 |
if generate_now and not st.session_state.get("busy"):
|
| 536 |
if not st.session_state.get("videos"):
|
| 537 |
st.error("No video loaded. Use 'Load Video' in the sidebar.")
|
|
|
|
| 546 |
if HAS_GENAI and genai is not None:
|
| 547 |
genai.configure(api_key=key_to_use)
|
| 548 |
except Exception:
|
| 549 |
+
logger.exception("genai configure failed")
|
| 550 |
|
| 551 |
model_id = model_input_value or st.session_state.get("preferred_model") or "gemini-2.5-flash-lite"
|
| 552 |
if st.session_state.get("last_model") != model_id:
|
| 553 |
st.session_state["last_model"] = ""
|
| 554 |
+
# no phi agent creation per plan
|
| 555 |
|
| 556 |
processed = st.session_state.get("processed_file")
|
| 557 |
current_path = st.session_state.get("videos")
|
|
|
|
| 568 |
if not HAS_GENAI:
|
| 569 |
raise RuntimeError("google.generativeai SDK not available; install it.")
|
| 570 |
local_path = current_path
|
| 571 |
+
|
| 572 |
+
# Decide whether to compress based on threshold (per plan ≤ threshold upload unchanged)
|
| 573 |
+
try:
|
| 574 |
+
file_size_mb = os.path.getsize(local_path) / (1024 * 1024)
|
| 575 |
+
except Exception:
|
| 576 |
+
file_size_mb = None
|
| 577 |
+
|
| 578 |
+
compressed = False
|
| 579 |
+
upload_path = local_path
|
| 580 |
+
threshold_mb = st.session_state.get("compression_threshold_mb", 200)
|
| 581 |
+
if file_size_mb is not None and file_size_mb > threshold_mb:
|
| 582 |
+
# compress with conservative settings; allow user to tune via constants if desired
|
| 583 |
+
compressed_path = str(Path(local_path).with_name(Path(local_path).stem + "_compressed.mp4"))
|
| 584 |
+
with st.spinner("Compressing video before upload..."):
|
| 585 |
+
upload_path = compress_video(local_path, compressed_path, crf=28, preset="fast")
|
| 586 |
+
if upload_path != local_path:
|
| 587 |
+
compressed = True
|
| 588 |
|
| 589 |
with st.spinner(f"Uploading video{' (compressed)' if compressed else ''}..."):
|
| 590 |
try:
|
|
|
|
| 622 |
max_tokens = 2048 if "2.5" in model_used else 1024
|
| 623 |
est_tokens = max_tokens
|
| 624 |
|
| 625 |
+
# Generate via Responses API
|
| 626 |
+
try:
|
| 627 |
+
gen_progress_placeholder = st.empty()
|
| 628 |
+
gen_status = gen_progress_placeholder.text("Starting generation...")
|
| 629 |
+
start_gen = time.time()
|
| 630 |
+
def gen_progress_cb(stage, elapsed, info):
|
| 631 |
+
try:
|
| 632 |
+
gen_status.text(f"Stage: {stage} — elapsed: {elapsed}s — {info}")
|
| 633 |
+
except Exception:
|
| 634 |
+
pass
|
| 635 |
+
out = generate_via_responses_api(prompt_text, processed, model_used, max_tokens=max_tokens, timeout=st.session_state.get("generation_timeout", 300), progress_callback=gen_progress_cb)
|
| 636 |
+
gen_progress_placeholder.text(f"Generation complete in {int(time.time()-start_gen)}s")
|
| 637 |
+
except Exception as e:
|
| 638 |
+
tb = traceback.format_exc()
|
| 639 |
+
st.session_state["last_error"] = f"Responses API error: {e}\n\nTraceback:\n{tb}"
|
| 640 |
+
st.error("An error occurred while generating the story. You can try Generate again; the uploaded video will be reused.")
|
| 641 |
+
out = ""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 642 |
|
| 643 |
if out:
|
| 644 |
out = remove_prompt_echo(prompt_text, out)
|
|
|
|
| 661 |
|
| 662 |
except Exception as e:
|
| 663 |
tb = traceback.format_exc()
|
| 664 |
+
st.session_state["last_error"] = f"{e}\n\nTraceback:\n{tb}"
|
| 665 |
st.error("An error occurred while generating the story. You can try Generate again; the uploaded video will be reused.")
|
| 666 |
finally:
|
| 667 |
st.session_state["busy"] = False
|
| 668 |
|
| 669 |
+
# Display existing analysis
|
| 670 |
if st.session_state.get("analysis_out"):
|
| 671 |
just_loaded_same = (st.session_state.get("last_loaded_path") == st.session_state.get("videos"))
|
| 672 |
if not just_loaded_same:
|
| 673 |
st.subheader("Analysis Result")
|
| 674 |
st.markdown(st.session_state.get("analysis_out"))
|
| 675 |
|
| 676 |
+
# Last error expander
|
| 677 |
if st.session_state.get("last_error"):
|
| 678 |
with st.expander("Last Error", expanded=False):
|
| 679 |
st.write(st.session_state.get("last_error"))
|