File size: 17,600 Bytes
6faf270 452ae25 175b0d0 dd37a0f a4208bd 175b0d0 a4208bd 33b5870 1c69ffa 452ae25 dd37a0f 1c69ffa 452ae25 1c69ffa 452ae25 33b5870 dd37a0f 175b0d0 dd37a0f 175b0d0 33b5870 a4208bd 175b0d0 a4208bd 175b0d0 dd37a0f a4208bd 452ae25 a4208bd f04e85a 452ae25 33b5870 452ae25 33b5870 6faf270 1c69ffa 6faf270 1c69ffa 6faf270 1c69ffa 6faf270 1c69ffa 452ae25 a4208bd 1c69ffa 175b0d0 452ae25 244b845 33b5870 175b0d0 a4208bd 175b0d0 a4208bd 452ae25 33b5870 a4208bd 244b845 6faf270 33b5870 a4208bd a5ce3c4 452ae25 a5ce3c4 452ae25 a5ce3c4 452ae25 a5ce3c4 3fc0626 452ae25 a4208bd 452ae25 a4208bd 1c69ffa a4208bd 452ae25 a4208bd 452ae25 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 | from typing import Optional, Tuple
from fastapi import FastAPI, UploadFile, File, Form
from fastapi.responses import FileResponse
from fastapi.middleware.cors import CORSMiddleware
from PIL import Image, ExifTags
import io
import hashlib
import httpx
import os
import base64
import json
import asyncio
import cv2
import tempfile
import fitz # pymupdf
import pypdf
# ---------------- API KEYS ----------------
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
HF_API_KEY = os.getenv("HF_API_KEY")
GROQ_API_URL = "https://api.groq.com/openai/v1/chat/completions"
ROBERTA_FAKE_NEWS_URL = "https://router.huggingface.co/hf-inference/models/hamzab/roberta-fake-news-classification"
ROBERTA_AI_TEXT_URL = "https://router.huggingface.co/hf-inference/models/openai-community/roberta-base-openai-detector"
# ---------------- APP ----------------
app = FastAPI()
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_methods=["*"],
allow_headers=["*"],
)
@app.get("/")
def read_root():
return FileResponse("index.html")
# ==============================
# UTILITIES
# ==============================
def calculate_sha256(contents: bytes):
return hashlib.sha256(contents).hexdigest()
def calculate_metadata_risk(image: Image.Image):
risk = 0.0
try:
exif = image._getexif()
if exif is None:
risk += 0.1
else:
for tag, value in exif.items():
decoded = ExifTags.TAGS.get(tag, tag)
if decoded == "Software":
risk += 0.2
except Exception:
risk += 0.1
return min(risk, 1.0)
def fusion_score(model_score: float, metadata_risk: float):
final = 0.9 * model_score + 0.1 * metadata_risk
authenticity = (1 - final) * 100
fake = final * 100
return authenticity, fake
def normalize_output(label_prob_dict: dict) -> float:
FAKE_KEYWORDS = ["fake", "ai", "generated", "manipulated", "deepfake", "artificial", "synthetic", "machine"]
REAL_KEYWORDS = ["real", "authentic", "genuine", "human", "original"]
fake_score = 0.0
uncertain_score = 0.0
for label, prob in label_prob_dict.items():
label_lower = label.lower()
if any(k in label_lower for k in FAKE_KEYWORDS):
fake_score += prob
elif any(k in label_lower for k in REAL_KEYWORDS):
pass
else:
uncertain_score += prob
fake_score += 0.4 * uncertain_score
return min(fake_score, 1.0)
def make_confidence(authenticity, fake):
diff = abs(authenticity - fake)
return "low" if diff < 20 else "medium" if diff < 40 else "high"
# ==============================
# GROQ VISION (images)
# ==============================
async def call_groq_vision(contents: bytes) -> Tuple[Optional[float], str]:
if not GROQ_API_KEY:
print("No GROQ_API_KEY set")
return None, ""
try:
base64_image = base64.b64encode(contents).decode('utf-8')
payload = {
"model": "meta-llama/llama-4-scout-17b-16e-instruct",
"messages": [
{
"role": "user",
"content": [
{
"type": "text",
"text": """You are a forensic image analyst expert. Analyze this image for signs of AI generation or manipulation.
Look for:
- Unnatural skin texture or too-perfect features
- Inconsistent lighting or shadows
- Background anomalies or blurring
- Artifacts typical of diffusion models (Midjourney, DALL-E, Stable Diffusion)
- Overly smooth or painterly textures
- Unnatural hair or eye details
- Signs of face swapping or deepfake manipulation
- EXIF/compression patterns typical of AI tools
Respond ONLY in this exact JSON format, nothing else:
{"fake_probability": 0.0, "reasoning": "brief reason"}
fake_probability must be between 0.0 (definitely real) and 1.0 (definitely AI/fake)."""
},
{
"type": "image_url",
"image_url": {
"url": f"data:image/jpeg;base64,{base64_image}"
}
}
]
}
],
"max_tokens": 200,
"temperature": 0.1
}
async with httpx.AsyncClient(timeout=30.0) as client:
response = await client.post(
GROQ_API_URL,
headers={
"Authorization": f"Bearer {GROQ_API_KEY}",
"Content-Type": "application/json"
},
json=payload
)
response.raise_for_status()
data = response.json()
text = data["choices"][0]["message"]["content"]
print(f"Groq vision response: {text}")
clean = text.strip().replace("```json", "").replace("```", "")
result = json.loads(clean)
return float(result["fake_probability"]), result.get("reasoning", "")
except Exception as e:
print(f"Groq vision failed: {e}")
return None, ""
# ==============================
# GROQ TEXT (for AI writing detection)
# ==============================
async def call_groq_text(text: str) -> Tuple[Optional[float], str]:
if not GROQ_API_KEY:
return None, ""
try:
payload = {
"model": "llama-3.3-70b-versatile",
"messages": [
{
"role": "user",
"content": f"""You are a forensic text analyst. Analyze the following text and determine if it is AI-generated or written by a human. Also check if it could be a forged government document or fake news.
Look for:
- Overly formal or repetitive sentence structure typical of LLMs
- Lack of personal voice or human inconsistencies
- Suspiciously perfect grammar with no natural errors
- Generic phrasing commonly used by AI models
- For government documents: inconsistent terminology, wrong formats, suspicious clauses
- For news: sensational language, lack of credible sources, misleading framing
Text to analyze:
\"\"\"
{text[:4000]}
\"\"\"
Respond ONLY in this exact JSON format, nothing else:
{{"fake_probability": 0.0, "reasoning": "brief reason"}}
fake_probability must be between 0.0 (definitely human/authentic) and 1.0 (definitely AI-generated/forged)."""
}
],
"max_tokens": 200,
"temperature": 0.1
}
async with httpx.AsyncClient(timeout=30.0) as client:
response = await client.post(
GROQ_API_URL,
headers={
"Authorization": f"Bearer {GROQ_API_KEY}",
"Content-Type": "application/json"
},
json=payload
)
response.raise_for_status()
data = response.json()
text_response = data["choices"][0]["message"]["content"]
print(f"Groq text response: {text_response}")
clean = text_response.strip().replace("```json", "").replace("```", "")
result = json.loads(clean)
return float(result["fake_probability"]), result.get("reasoning", "")
except Exception as e:
print(f"Groq text failed: {e}")
return None, ""
# ==============================
# ROBERTA (for fake news + AI text)
# ==============================
async def call_roberta(url: str, text: str, name: str) -> Optional[float]:
if not HF_API_KEY:
print(f"No HF_API_KEY, skipping {name}")
return None
try:
async with httpx.AsyncClient(timeout=30.0) as client:
response = await client.post(
url,
headers={"Authorization": f"Bearer {HF_API_KEY}"},
json={"inputs": text[:512]}
)
response.raise_for_status()
data = response.json()
print(f"{name} response: {data}")
label_prob_dict = {item["label"]: item["score"] for item in data[0]}
return normalize_output(label_prob_dict)
except Exception as e:
print(f"{name} failed: {e}")
return None
# ==============================
# ANALYZERS
# ==============================
async def analyze_image(contents: bytes, content_type: str = "image/jpeg"):
image = Image.open(io.BytesIO(contents)).convert("RGB")
if len(contents) > 20 * 1024 * 1024:
print("Image too large for Groq")
score, reasoning = None, "Image too large for analysis"
else:
score, reasoning = await call_groq_vision(contents)
combined_model_score = score if score is not None else 0.5
models_used = ["Groq_Llama4"] if score is not None else []
metadata_risk = calculate_metadata_risk(image)
authenticity, fake = fusion_score(combined_model_score, metadata_risk)
return {
"type": "image",
"authenticity": round(authenticity, 2),
"fake": round(fake, 2),
"confidence_level": make_confidence(authenticity, fake),
"models_used": models_used,
"details": {
"groq_score": round(score, 4) if score is not None else "unavailable",
"groq_reasoning": reasoning,
"metadata_risk": round(metadata_risk, 4),
}
}
async def analyze_video(contents: bytes):
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp4") as f:
f.write(contents)
tmp_path = f.name
try:
cap = cv2.VideoCapture(tmp_path)
frame_count = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
fps = cap.get(cv2.CAP_PROP_FPS)
duration = round(frame_count / fps, 1) if fps > 0 else 0
sample_indices = [int(frame_count * i / 5) for i in range(5)]
frames = []
for idx in sample_indices:
cap.set(cv2.CAP_PROP_POS_FRAMES, idx)
ret, frame = cap.read()
if ret:
pil_img = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
buf = io.BytesIO()
pil_img.save(buf, format="JPEG", quality=85)
frames.append(buf.getvalue())
cap.release()
os.unlink(tmp_path)
if not frames:
return {
"type": "video",
"authenticity": 50.0,
"fake": 50.0,
"confidence_level": "low",
"models_used": [],
"details": {
"groq_score": "unavailable",
"groq_reasoning": "Could not extract frames from video.",
"metadata_risk": 0.0,
"frames_analyzed": 0,
"video_duration": duration
}
}
scores = []
reasonings = []
for i, frame_bytes in enumerate(frames):
print(f"Analyzing frame {i+1}/{len(frames)}")
score, reasoning = await call_groq_vision(frame_bytes)
if score is not None:
scores.append(score)
reasonings.append(f"Frame {i+1}: {reasoning}")
if i < len(frames) - 1:
await asyncio.sleep(2)
combined_model_score = sum(scores) / len(scores) if scores else 0.5
models_used = ["Groq_Llama4"] if scores else []
groq_reasoning = " | ".join(reasonings) if reasonings else "All frame analyses failed."
authenticity = round((1 - combined_model_score) * 100, 2)
fake = round(combined_model_score * 100, 2)
return {
"type": "video",
"authenticity": authenticity,
"fake": fake,
"confidence_level": make_confidence(authenticity, fake),
"models_used": models_used,
"details": {
"groq_score": round(combined_model_score, 4),
"groq_reasoning": groq_reasoning,
"metadata_risk": 0.0,
"frames_analyzed": len(scores),
"video_duration": duration
}
}
except Exception as e:
print(f"Video analysis failed: {e}")
if os.path.exists(tmp_path):
os.unlink(tmp_path)
return {
"type": "video",
"authenticity": 50.0,
"fake": 50.0,
"confidence_level": "low",
"models_used": [],
"details": {
"groq_score": "unavailable",
"groq_reasoning": f"Analysis failed: {str(e)}",
"metadata_risk": 0.0,
"frames_analyzed": 0,
"video_duration": 0
}
}
async def analyze_text(text: str):
# all 3 in parallel
results = await asyncio.gather(
call_roberta(ROBERTA_FAKE_NEWS_URL, text, "RoBERTa_FakeNews"),
call_roberta(ROBERTA_AI_TEXT_URL, text, "RoBERTa_AIDetector"),
call_groq_text(text)
)
score1 = results[0]
score2 = results[1]
score3, reasoning = results[2]
scores = [(s, n) for s, n in [
(score1, "RoBERTa_FakeNews"),
(score2, "RoBERTa_AIDetector"),
(score3, "Groq_Llama3")
] if s is not None]
combined = sum(s for s, _ in scores) / len(scores) if scores else 0.5
models_used = [n for _, n in scores]
authenticity = round((1 - combined) * 100, 2)
fake = round(combined * 100, 2)
return {
"type": "text",
"authenticity": authenticity,
"fake": fake,
"confidence_level": make_confidence(authenticity, fake),
"models_used": models_used,
"details": {
"groq_score": round(score3, 4) if score3 is not None else "unavailable",
"roberta_fakenews_score": round(score1, 4) if score1 is not None else "unavailable",
"roberta_aidetector_score": round(score2, 4) if score2 is not None else "unavailable",
"groq_reasoning": reasoning,
"metadata_risk": 0.0,
}
}
async def analyze_pdf(contents: bytes):
scores = []
reasonings = []
try:
# extract text
reader = pypdf.PdfReader(io.BytesIO(contents))
full_text = ""
for page in reader.pages:
full_text += page.extract_text() or ""
if full_text.strip():
print(f"Extracted {len(full_text)} chars from PDF")
text_results = await asyncio.gather(
call_roberta(ROBERTA_FAKE_NEWS_URL, full_text, "RoBERTa_FakeNews"),
call_roberta(ROBERTA_AI_TEXT_URL, full_text, "RoBERTa_AIDetector"),
call_groq_text(full_text)
)
s1 = text_results[0]
s2 = text_results[1]
s3, text_reasoning = text_results[2]
if s1 is not None:
scores.append(s1)
reasonings.append(f"RoBERTa FakeNews: {round(s1*100)}% fake")
if s2 is not None:
scores.append(s2)
reasonings.append(f"RoBERTa AI Detector: {round(s2*100)}% AI-generated")
if s3 is not None:
scores.append(s3)
reasonings.append(f"Groq text: {text_reasoning}")
# extract and analyze images inside PDF
doc = fitz.open(stream=contents, filetype="pdf")
image_count = 0
for page in doc:
for img in page.get_images():
if image_count >= 3:
break
xref = img[0]
base_image = doc.extract_image(xref)
img_bytes = base_image["image"]
await asyncio.sleep(2)
img_score, img_reasoning = await call_groq_vision(img_bytes)
if img_score is not None:
scores.append(img_score)
reasonings.append(f"Image {image_count+1}: {img_reasoning}")
image_count += 1
doc.close()
except Exception as e:
print(f"PDF analysis error: {e}")
combined = sum(scores) / len(scores) if scores else 0.5
models_used = ["RoBERTa_FakeNews", "RoBERTa_AIDetector", "Groq_Llama3+Vision"] if scores else []
authenticity = round((1 - combined) * 100, 2)
fake = round(combined * 100, 2)
return {
"type": "pdf",
"authenticity": authenticity,
"fake": fake,
"confidence_level": make_confidence(authenticity, fake),
"models_used": models_used,
"details": {
"groq_score": "see breakdown",
"groq_reasoning": " | ".join(reasonings) if reasonings else "No content extracted",
"metadata_risk": 0.0,
}
}
# ==============================
# ROUTER
# ==============================
@app.post("/analyze")
async def analyze(
file: Optional[UploadFile] = File(None),
text: Optional[str] = Form(None)
):
# plain text input
if text and not file:
result = await analyze_text(text)
result["sha256"] = hashlib.sha256(text.encode()).hexdigest()
return result
if not file:
return {"error": "No file or text provided"}
contents = await file.read()
sha256 = calculate_sha256(contents)
if file.content_type.startswith("image/"):
result = await analyze_image(contents, file.content_type)
elif file.content_type.startswith("video/"):
result = await analyze_video(contents)
elif file.content_type == "application/pdf":
result = await analyze_pdf(contents)
else:
return {"error": "Unsupported file type"}
result["sha256"] = sha256
return result
|