Wave 2: Script quality and other changes

README.md

@@ -4,14 +4,17 @@ AI-powered tool for generating YouTube Shorts / TikTok videos with script genera
 
 ## Features
 
-- **AI Script Generation** — viral-optimized scripts in Brazilian Portuguese via Pollinations AI
-- **AI Image Generation** — background images via Pollinations AI (images-only mode)
+- **AI Script Generation** — curiosity-driven scripts in Brazilian Portuguese via Pollinations AI with fact verification and hallucination guards
+- **Web-Grounded Scripts** — automatic web search generates independent queries, grounds the script in real sources, then cross-checks every claim
+- **Title Validation** — auto-validates hashtag count (3+), length (≤ 100 chars), and lowercases tags
+- **AI Image Generation** — background images via Pollinations AI or **real web images** via DuckDuckGo search (default), with NSFW domain/keyword filter
 - **Text-to-Speech** — natural audio via Edge TTS
 - **Subtitle System** — VTT generation + word-level highlight rendering
-- **Video Composition** — blurred YouTube background or AI images with smooth overlay animation
+- **Video Composition** — blurred YouTube background or AI/web images with smooth overlay animation
 - **YouTube Integration** — download any video as background footage
-- **Two pipelines**: normal (YouTube bg + optional AI image overlays) and images-only (AI images + overlay animation, no YouTube bg)
+- **Two pipelines**: normal (YouTube bg + optional image overlays) and images-only (AI/web images + overlay animation, no YouTube bg)
 - **Typer CLI** — nested subcommands, auto-generated `--help`, shell completion
+- **Batch Processing** — semicolon-separated subjects for multi-video runs
 
 ## Installation
 
@@ -37,7 +40,7 @@ uv pip install -e ".[dev]"
 # See available commands
 autoshorts --help
 
-# Generate an explainer video from a topic
+# Generate an explainer video from a topic (web search + web images by default)
 autoshorts new explainer "artificial intelligence"
 
 # AI images only (no YouTube background)
@@ -46,14 +49,20 @@ autoshorts new explainer "space exploration" --images-only
 # Use a YouTube video as background footage
 autoshorts new explainer --youtube-url "https://youtube.com/watch?v=VIDEO_ID"
 
-# Skip AI image overlays (blurred bg only)
+# Skip image overlays (blurred bg only)
 autoshorts new explainer "climate change" --no-images
 
-# Batch mode
-autoshorts new explainer --batch "robotics" "quantum computing" "neural networks"
+# Batch mode with semicolon-separated subjects
+autoshorts new explainer --batch "robotics; quantum computing; neural networks"
 
-# Web search for richer script content
-autoshorts new explainer "oceanography" --web-search
+# Image source: 'ai' uses Pollinations (default is 'web' via DDGS)
+autoshorts new explainer "oceanography" --images ai
+
+# Script tone: 'corporate' (neutral, factual) or 'opinionated' (curiosity-driven, narrative)
+autoshorts new explainer "bitcoin" --tone corporate
+
+# Disable web search (uses model knowledge only)
+autoshorts new explainer "neural networks" --no-web-search
 
 # Auto-shutdown after completion
 autoshorts new explainer "future technology" --goodnight
@@ -102,23 +111,29 @@ AutoShorts/
 │       │   └── explainer.py        # ExplainerGenerator (both pipelines)
 │       └── modules/                # Core modules
 │           ├── config.py
+│           ├── image_searcher.py   # Web/AI image search + NSFW filter
 │           ├── logging_system.py
-│           ├── script_generator.py
+│           ├── script_generator.py # Script gen, fact verification, title validation
 │           ├── subtitle_system.py
 │           ├── tts_system.py
 │           ├── utils.py
-│           ├── video_background.py
-│           └── video_compositor.py
+│           ├── video_background.py # YouTube search & download
+│           ├── video_compositor.py
+│           └── web_search.py       # DuckDuckGo web search
 ├── tests/
-│   ├── test_cli.py                 # CLI layer (28 tests)
+│   ├── test_cli.py                 # CLI layer (32 tests)
+│   ├── test_config.py
+│   ├── test_edge_cases.py
 │   ├── test_fluximages.py          # Explainer generator tests
-│   ├── test_video_background.py    # Video background (24 tests)
-│   ├── test_video_compositor.py    # Video compositor (11 tests)
+│   ├── test_init.py
+│   ├── test_integration.py
 │   ├── test_script_generator.py
 │   ├── test_subtitle_system.py
-│   ├── test_utils.py
-│   ├── test_edge_cases.py
 │   ├── test_tts_system.py
+│   ├── test_utils.py
+│   ├── test_video_background.py    # Video background (24 tests)
+│   ├── test_video_compositor.py    # Video compositor (11 tests)
+│   ├── test_web_search.py
 │   └── conftest.py
 ├── fonts/                          # Bundled Bebas Neue font
 ├── .env.example
@@ -134,13 +149,15 @@ Core:
 - `edge-tts` — text-to-speech
 - `requests` — HTTP client
 - `yt-dlp` — YouTube downloading
+- `duckduckgo-search` — web search and image search (DDGS)
+- `Pillow` — image processing and resizing
 - `webvtt-py` — subtitle processing
 - `python-dotenv` — environment loading
 - `typer` — CLI framework
 
 Dev:
 - `pytest` + `pytest-asyncio` + `pytest-cov`
-- `black` + `ruff` + `mypy`
+- `ruff` + `mypy`
 
 ## Development
 

src/autoshorts/cli/commands/explainer.py

@@ -1,4 +1,5 @@
 import asyncio
+import re
 import time
 from pathlib import Path
 
@@ -21,7 +22,7 @@ def explainer_command(
     goodnight: bool = typer.Option(
         False, "--goodnight", help="Shutdown after processing"
     ),
-    batch: list[str] = typer.Option(None, "--batch", help="Batch: multiple subjects"),
+    batch: str = typer.Option(None, "--batch", help="Batch: semicolon-separated subjects (e.g. 'topic1; topic2')"),
     no_web_search: bool = typer.Option(
         False, "--no-web-search", help="Disable web search (use model knowledge only)"
     ),
@@ -46,7 +47,7 @@ def explainer_command(
 
     subjects: list[str | None] = []
     if batch:
-        subjects = [s for s in batch]
+        subjects = [s.strip() for s in batch.split(";") if s.strip()]
     elif subject:
         subjects = [subject]
     elif youtube_url:
@@ -56,6 +57,7 @@ def explainer_command(
     output_path = Path(output)
     success_count = 0
     total_count = len(subjects)
+    def _sanitize(s): return re.sub(r'[\\/*?:"<>|]', "", s).replace(" ", "_")[:20]
 
     for i, subj in enumerate(subjects, 1):
         log(f"Processing {i}/{total_count}: {subj or 'youtube-url'}")
@@ -64,14 +66,14 @@ def explainer_command(
                 out_dir = output_path.parent
                 if is_batch:
                     prefix = "explainer_" if images_only else "as_"
-                    name = f"{prefix}{subj.replace(' ', '_')[:20] if subj else 'video'}_{int(time.time())}.mp4"
+                    name = f"{prefix}{_sanitize(subj) if subj else 'video'}_{int(time.time())}.mp4"
                 else:
                     name = output_path.name
             else:
                 out_dir = output_path
                 if is_batch:
                     prefix = "explainer_" if images_only else "as_"
-                    name = f"{prefix}{subj.replace(' ', '_')[:20] if subj else 'video'}_{int(time.time())}.mp4"
+                    name = f"{prefix}{_sanitize(subj) if subj else 'video'}_{int(time.time())}.mp4"
                 else:
                     prefix = "explainer_" if images_only else "autoshorts_"
                     name = f"{prefix}{int(time.time())}.mp4"

src/autoshorts/generators/explainer.py

@@ -104,6 +104,12 @@ async def _run_normal_pipeline(self) -> bool:
                 script = self.script_generator.generate_script(subject)
             else:
                 script = self.script_generator.generate_script_from_metadata(title, "")
+            if not script or len(script) < 3:
+                log(
+                    f"Script generation failed: got {len(script)} paragraphs, need >= 3",
+                    "ERROR",
+                )
+                return False
             log(f"Generated script with {len(script)} paragraphs")
 
             log("Step 3: Generating TTS audio...")
@@ -226,6 +232,12 @@ async def _run_images_only_pipeline(self) -> bool:
             paragraphs, _ = self.script_generator.generate_script_with_prompts(
                 self.subject
             )
+            if not paragraphs or len(paragraphs) < 3:
+                log(
+                    f"Script generation failed: got {len(paragraphs)} paragraphs, need >= 3",
+                    "ERROR",
+                )
+                return False
 
             log("Step 2: Generating TTS audio...")
             audio_path = await self.tts_system.generate_audio_only(
@@ -315,7 +327,8 @@ def apply_opacity(get_frame, t):
             return np.minimum(255, frame * opacity).astype("uint8")
 
         clip = clip.with_effects([vfx.Resize(scale_anim)])
-        return clip.transform(apply_opacity)
+        clip = clip.transform(apply_opacity)
+        return clip.with_position(("center", "center"))
 
     def _create_flux_video(
         self, img_paths: list, audio_path: str, paragraphs: list, output_path: str

src/autoshorts/modules/image_searcher.py

@@ -1,5 +1,6 @@
 import hashlib
 import random
+import re
 from pathlib import Path
 from urllib.parse import quote
 
@@ -21,6 +22,18 @@
 )
 from .logging_system import log
 
+BLOCKED_DOMAINS: set[str] = {
+    "crossdresser", "sissy", "femboy", "hentai", "rule34",
+    "xvideos", "xnxx", "xhamster", "pornhub", "onlyfans",
+    "redtube", "youporn", "erotic", "nsfw",
+}
+
+BLOCKED_KEYWORDS: set[str] = {
+    "crossdresser", "sissy", "femboy", "hentai", "rule34",
+    "nsfw", "xxx", "18+", "erotic",
+    "nude", "naked", "seductive",
+}
+
 
 class ImageSearcher:
     def __init__(
@@ -40,6 +53,21 @@ def __init__(
         self.max_per_query = max_per_query
         IMAGE_CACHE_DIR.mkdir(parents=True, exist_ok=True)
 
+    @staticmethod
+    def _is_nsfw(result: dict) -> bool:
+        text = (
+            f"{result.get('image') or ''} "
+            f"{result.get('url') or result.get('source') or ''} "
+            f"{result.get('title') or ''}"
+        )
+        for d in BLOCKED_DOMAINS:
+            if re.search(rf"(?:^|[\W_]){re.escape(d)}(?:$|[\W_])", text, re.IGNORECASE):
+                return True
+        for kw in BLOCKED_KEYWORDS:
+            if re.search(rf"(?:^|[\W_]){re.escape(kw)}(?:$|[\W_])", text, re.IGNORECASE):
+                return True
+        return False
+
     def search_images(self, query: str) -> list[dict]:
         try:
             from ddgs import DDGS
@@ -115,6 +143,7 @@ def get_images(self, prompts: list[str]) -> list[str]:
                 continue
 
             results = self.search_images(prompt)
+            results = [r for r in results if not self._is_nsfw(r)]
             downloaded = False
             for r in results:
                 url = r.get("image", "")