vidwise 0.1.0__tar.gz

vidwise-0.1.0/.claude-plugin/marketplace.json

@@ -0,0 +1,16 @@
+{
+  "name": "vidwise",
+  "description": "LLMs can't watch videos. vidwise gives them eyes. Extract transcripts, frames, and visual guides from any video.",
+  "owner": {
+    "name": "Juan Pablo Djeredjian",
+    "email": "jpdjeredjian@gmail.com"
+  },
+  "plugins": [
+    {
+      "name": "vidwise",
+      "source": "./plugin",
+      "description": "Make any video AI-readable. Extract transcripts, frames, and visual guides from videos using Claude Code's native multimodal AI — no API key needed.",
+      "version": "0.1.0"
+    }
+  ]
+}

vidwise-0.1.0/.github/workflows/ci.yml

@@ -0,0 +1,31 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - run: pip install ruff
+      - run: ruff check src/
+
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+      - run: pip install -e ".[dev]"
+      - run: pytest

vidwise-0.1.0/.github/workflows/publish.yml

@@ -0,0 +1,20 @@
+name: Publish to PyPI
+
+on:
+  push:
+    tags:
+      - "v*"
+
+jobs:
+  publish:
+    runs-on: ubuntu-latest
+    permissions:
+      id-token: write
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+      - run: pip install build
+      - run: python -m build
+      - uses: pypa/gh-action-pypi-publish@release/v1

vidwise-0.1.0/.gitignore

@@ -0,0 +1,23 @@
+__pycache__/
+*.py[cod]
+*$py.class
+*.egg-info/
+dist/
+build/
+.eggs/
+*.egg
+.venv/
+venv/
+env/
+.mypy_cache/
+.ruff_cache/
+.pytest_cache/
+*.wav
+*.mp4
+*.webm
+*.mkv
+*.avi
+*.mov
+output-*/
+vidwise-output-*/
+TODO.md

vidwise-0.1.0/CHANGELOG.md

@@ -0,0 +1,19 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
+and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.1.0] - 2026-02-26
+
+### Added
+
+- Core CLI with `vidwise <source>` command
+- Whisper-powered transcription (`.txt`, `.srt`, `.json` outputs)
+- Frame extraction every N seconds with timestamp-based naming
+- Smart key frame selection via pixel-difference analysis
+- URL support via yt-dlp (YouTube, Loom, 1000+ sites)
+- AI-powered visual guide generation (Claude and OpenAI providers)
+- Claude Code plugin with `/vidwise` slash command
+- Parallel frame analysis via Claude Code subagents (no API key needed)

vidwise-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Juan Pablo Djeredjian
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

vidwise-0.1.0/PKG-INFO

@@ -0,0 +1,240 @@
+Metadata-Version: 2.4
+Name: vidwise
+Version: 0.1.0
+Summary: LLMs can't watch videos. vidwise gives them eyes.
+Project-URL: Homepage, https://github.com/jpdjere/vidwise
+Project-URL: Repository, https://github.com/jpdjere/vidwise
+Project-URL: Issues, https://github.com/jpdjere/vidwise/issues
+Author: Juan Pablo Djeredjian
+License-Expression: MIT
+License-File: LICENSE
+Keywords: ai,frames,knowledge,llm,transcript,video,whisper
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Multimedia :: Video
+Requires-Python: >=3.10
+Requires-Dist: anthropic>=0.40
+Requires-Dist: click>=8.0
+Requires-Dist: openai-whisper>=20231117
+Requires-Dist: openai>=1.0
+Requires-Dist: yt-dlp>=2023.0
+Provides-Extra: dev
+Requires-Dist: pytest-cov; extra == 'dev'
+Requires-Dist: pytest>=7.0; extra == 'dev'
+Requires-Dist: ruff; extra == 'dev'
+Provides-Extra: fast
+Requires-Dist: faster-whisper>=1.0; extra == 'fast'
+Description-Content-Type: text/markdown
+
+<p align="center">
+  <img src="assets/banner.png" alt="vidwise — LLMs can't watch videos. vidwise gives them eyes." width="700">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/vidwise/"><img src="https://img.shields.io/pypi/v/vidwise?color=blue" alt="PyPI"></a>
+  <a href="https://pypi.org/project/vidwise/"><img src="https://img.shields.io/pypi/pyversions/vidwise" alt="Python"></a>
+  <a href="https://github.com/jpdjere/vidwise/blob/main/LICENSE"><img src="https://img.shields.io/github/license/jpdjere/vidwise" alt="License"></a>
+  <a href="https://github.com/jpdjere/vidwise/actions/workflows/ci.yml"><img src="https://github.com/jpdjere/vidwise/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
+</p>
+
+---
+
+Videos are the biggest blind spot for AI. A 5-minute Loom bug report, a 30-minute tutorial, a conference talk — all completely opaque to your LLM. You either watch the whole thing yourself or lose the knowledge.
+
+**vidwise** extracts the visual and audio knowledge from any video into structured, LLM-consumable markdown. Feed the output to any LLM and it instantly "understands" the video.
+
+```
+Video ─→ vidwise ─→ Transcript + Key Frames + Visual Guide ─→ LLM Context
+```
+
+## What can you do with it?
+
+| Scenario | What happens |
+|----------|-------------|
+| **Debug a Loom bug report** | Feed the output to Claude → it "sees" the bug, the UI state, the error messages |
+| **Absorb a tutorial** | 30-min coding video → structured knowledge your LLM can answer questions about |
+| **Process a meeting** | Extract decisions, action items, and what was on screen |
+| **Learn from a talk** | Turn any conference presentation into searchable, queryable knowledge |
+| **Onboard faster** | Training videos become AI-queryable — new hires get instant answers |
+
+## Why vidwise?
+
+| | |
+|---|---|
+| **See the whole picture** | Most tools only extract audio. vidwise captures both what was *said* and what was *shown* — UI states, error messages, slides, code, diagrams. |
+| **Process once, query forever** | The output is a self-contained artifact. Feed it to any LLM, any number of times, at zero additional cost. No re-uploading, no re-processing. |
+| **Works with any LLM** | Standard markdown + images. Claude, GPT, Gemini, Llama, Mistral — whatever you use. No vendor lock-in. |
+| **Your video stays local** | Whisper and ffmpeg run on your machine. Nothing leaves your computer unless you opt into AI guide generation. |
+| **Smart, not brute-force** | Pixel-difference analysis keeps only frames where the visual content actually changed. Less noise, better LLM understanding. |
+| **Human-readable AND machine-readable** | The output isn't just for LLMs — `guide.md` is a visual walkthrough you can read, share, and bookmark. One command, two audiences. |
+| **One command** | `vidwise recording.mp4` → transcript, key frames, and visual guide in a single portable directory. |
+
+> **Not just for LLMs.** The visual guide vidwise generates is a fully readable document with embedded screenshots — open it in VS Code, Obsidian, or GitHub and you have a skimmable walkthrough of the entire video. Share it with your team, bookmark it for later, or feed it to any LLM. One artifact, two audiences.
+
+## Quick Start
+
+```bash
+# Install
+pip install vidwise
+
+# Process a local video
+vidwise recording.mp4
+
+# Process a YouTube video
+vidwise https://youtube.com/watch?v=abc
+
+# With AI-powered visual guide
+export ANTHROPIC_API_KEY=sk-...   # or OPENAI_API_KEY
+vidwise recording.mp4 --provider claude
+```
+
+### Prerequisites
+
+- **Python 3.10+**
+- **ffmpeg** — `brew install ffmpeg` (macOS) or `apt install ffmpeg` (Linux)
+
+> **Lighter install?** `pip install "vidwise[fast]"` uses faster-whisper (~200MB) instead of openai-whisper (~2GB). 3-4x faster transcription, but without Apple Metal GPU support. vidwise auto-detects which backend is installed.
+
+## Usage
+
+```bash
+vidwise <source> [options]
+```
+
+| Option | Default | Description |
+|--------|---------|-------------|
+| `--model`, `-m` | `medium` | Whisper model: `tiny`, `base`, `small`, `medium`, `large` |
+| `--output-dir`, `-o` | auto | Output directory path |
+| `--no-guide` | off | Skip AI guide generation |
+| `--provider`, `-p` | `auto` | AI provider: `auto`, `claude`, `openai` |
+| `--frame-interval` | `2` | Seconds between frame captures |
+| `--frame-threshold` | `0.05` | Pixel diff threshold for key frame selection |
+
+### Examples
+
+```bash
+# Fast transcription of a short video
+vidwise demo.mp4 --model tiny --no-guide
+
+# YouTube tutorial with Claude-powered guide
+vidwise https://youtube.com/watch?v=abc --model small --provider claude
+
+# Loom bug report — default settings
+vidwise https://loom.com/share/abc123def
+```
+
+## Output
+
+vidwise creates a single self-contained directory:
+
+```
+vidwise-abc123-2026-02-26/
+├── video.mp4              # Source video
+├── audio.wav              # Extracted audio (16kHz mono)
+├── transcript.txt         # Plain text transcript
+├── transcript.srt         # Timestamped subtitles
+├── transcript.json        # Full Whisper output with segments
+├── frames/                # Key frames every 2 seconds
+│   ├── frame_0m00s.png
+│   ├── frame_0m02s.png
+│   ├── frame_0m04s.png
+│   └── ...
+└── guide.md               # Visual guide with embedded frames (if AI enabled)
+```
+
+The `guide.md` uses relative image paths — open it in any markdown viewer (VS Code, GitHub, Obsidian) and the images render inline.
+
+## How It Works
+
+```
+┌─────────────┐
+│  Video URL   │──→ yt-dlp download
+│  or local    │
+└──────┬───────┘
+       │
+       ▼
+┌──────────────┐     ┌──────────────────┐
+│   ffmpeg     │────→│  audio.wav       │──→ Whisper ──→ transcript.*
+│  (parallel)  │     │  (16kHz mono)    │
+│              │────→│  frames/         │──→ Key frame selection
+│              │     │  (every 2 sec)   │    (pixel diff filtering)
+└──────────────┘     └──────────────────┘
+                              │
+                              ▼
+                     ┌──────────────────┐
+                     │  AI Analysis     │  Claude API, OpenAI API,
+                     │  (optional)      │  or Claude Code (free)
+                     └────────┬─────────┘
+                              │
+                              ▼
+                     ┌──────────────────┐
+                     │   guide.md       │  Structured markdown with
+                     │                  │  embedded frame images
+                     └──────────────────┘
+```
+
+**Smart frame selection:** Not every frame matters. vidwise compares consecutive frames using pixel-difference analysis and only keeps frames where the visual content actually changed. A 10-minute video might have 300 raw frames but only ~40 meaningful ones.
+
+## Claude Code Plugin
+
+If you use [Claude Code](https://docs.anthropic.com/en/docs/claude-code), install vidwise as a plugin for **AI-powered guide generation without needing an API key** — Claude Code's native multimodal AI handles the analysis:
+
+```bash
+# Add the vidwise marketplace and install the plugin
+/plugin marketplace add jpdjere/vidwise
+/plugin install vidwise@vidwise
+
+# Then use it:
+/vidwise:vidwise recording.mp4
+/vidwise:vidwise https://loom.com/share/abc123
+```
+
+For local development or testing, you can also load directly:
+
+```bash
+claude --plugin-dir /path/to/vidwise/plugin
+```
+
+The plugin runs `vidwise --no-guide` for extraction, then uses Claude Code's built-in vision capabilities to analyze frames in parallel — completely free, no API key needed.
+
+## Whisper Model Sizes
+
+| Model | Speed | Quality | Best For |
+|-------|-------|---------|----------|
+| `tiny` | ~1 min/min | Basic | Quick tests, long videos |
+| `base` | ~2 min/min | Good | Short videos |
+| `small` | ~4 min/min | Better | Videos >30 min |
+| `medium` | ~8 min/min | Recommended | Default for most content |
+| `large` | ~16 min/min | Best | When accuracy is critical |
+
+*Speed estimates on Apple M-series. First run downloads model weights (one-time).*
+
+## Contributing
+
+Contributions are welcome! Please open an issue first to discuss what you'd like to change.
+
+```bash
+# Development setup
+git clone https://github.com/jpdjere/vidwise
+cd vidwise
+python -m venv .venv
+source .venv/bin/activate
+pip install -e ".[dev]"
+
+# Run tests
+pytest
+
+# Lint
+ruff check src/
+```
+
+## License
+
+[MIT](LICENSE)

vidwise-0.1.0/README.md

@@ -0,0 +1,205 @@
+<p align="center">
+  <img src="assets/banner.png" alt="vidwise — LLMs can't watch videos. vidwise gives them eyes." width="700">
+</p>
+
+<p align="center">
+  <a href="https://pypi.org/project/vidwise/"><img src="https://img.shields.io/pypi/v/vidwise?color=blue" alt="PyPI"></a>
+  <a href="https://pypi.org/project/vidwise/"><img src="https://img.shields.io/pypi/pyversions/vidwise" alt="Python"></a>
+  <a href="https://github.com/jpdjere/vidwise/blob/main/LICENSE"><img src="https://img.shields.io/github/license/jpdjere/vidwise" alt="License"></a>
+  <a href="https://github.com/jpdjere/vidwise/actions/workflows/ci.yml"><img src="https://github.com/jpdjere/vidwise/actions/workflows/ci.yml/badge.svg" alt="CI"></a>
+</p>
+
+---
+
+Videos are the biggest blind spot for AI. A 5-minute Loom bug report, a 30-minute tutorial, a conference talk — all completely opaque to your LLM. You either watch the whole thing yourself or lose the knowledge.
+
+**vidwise** extracts the visual and audio knowledge from any video into structured, LLM-consumable markdown. Feed the output to any LLM and it instantly "understands" the video.
+
+```
+Video ─→ vidwise ─→ Transcript + Key Frames + Visual Guide ─→ LLM Context
+```
+
+## What can you do with it?
+
+| Scenario | What happens |
+|----------|-------------|
+| **Debug a Loom bug report** | Feed the output to Claude → it "sees" the bug, the UI state, the error messages |
+| **Absorb a tutorial** | 30-min coding video → structured knowledge your LLM can answer questions about |
+| **Process a meeting** | Extract decisions, action items, and what was on screen |
+| **Learn from a talk** | Turn any conference presentation into searchable, queryable knowledge |
+| **Onboard faster** | Training videos become AI-queryable — new hires get instant answers |
+
+## Why vidwise?
+
+| | |
+|---|---|
+| **See the whole picture** | Most tools only extract audio. vidwise captures both what was *said* and what was *shown* — UI states, error messages, slides, code, diagrams. |
+| **Process once, query forever** | The output is a self-contained artifact. Feed it to any LLM, any number of times, at zero additional cost. No re-uploading, no re-processing. |
+| **Works with any LLM** | Standard markdown + images. Claude, GPT, Gemini, Llama, Mistral — whatever you use. No vendor lock-in. |
+| **Your video stays local** | Whisper and ffmpeg run on your machine. Nothing leaves your computer unless you opt into AI guide generation. |
+| **Smart, not brute-force** | Pixel-difference analysis keeps only frames where the visual content actually changed. Less noise, better LLM understanding. |
+| **Human-readable AND machine-readable** | The output isn't just for LLMs — `guide.md` is a visual walkthrough you can read, share, and bookmark. One command, two audiences. |
+| **One command** | `vidwise recording.mp4` → transcript, key frames, and visual guide in a single portable directory. |
+
+> **Not just for LLMs.** The visual guide vidwise generates is a fully readable document with embedded screenshots — open it in VS Code, Obsidian, or GitHub and you have a skimmable walkthrough of the entire video. Share it with your team, bookmark it for later, or feed it to any LLM. One artifact, two audiences.
+
+## Quick Start
+
+```bash
+# Install
+pip install vidwise
+
+# Process a local video
+vidwise recording.mp4
+
+# Process a YouTube video
+vidwise https://youtube.com/watch?v=abc
+
+# With AI-powered visual guide
+export ANTHROPIC_API_KEY=sk-...   # or OPENAI_API_KEY
+vidwise recording.mp4 --provider claude
+```
+
+### Prerequisites
+
+- **Python 3.10+**
+- **ffmpeg** — `brew install ffmpeg` (macOS) or `apt install ffmpeg` (Linux)
+
+> **Lighter install?** `pip install "vidwise[fast]"` uses faster-whisper (~200MB) instead of openai-whisper (~2GB). 3-4x faster transcription, but without Apple Metal GPU support. vidwise auto-detects which backend is installed.
+
+## Usage
+
+```bash
+vidwise <source> [options]
+```
+
+| Option | Default | Description |
+|--------|---------|-------------|
+| `--model`, `-m` | `medium` | Whisper model: `tiny`, `base`, `small`, `medium`, `large` |
+| `--output-dir`, `-o` | auto | Output directory path |
+| `--no-guide` | off | Skip AI guide generation |
+| `--provider`, `-p` | `auto` | AI provider: `auto`, `claude`, `openai` |
+| `--frame-interval` | `2` | Seconds between frame captures |
+| `--frame-threshold` | `0.05` | Pixel diff threshold for key frame selection |
+
+### Examples
+
+```bash
+# Fast transcription of a short video
+vidwise demo.mp4 --model tiny --no-guide
+
+# YouTube tutorial with Claude-powered guide
+vidwise https://youtube.com/watch?v=abc --model small --provider claude
+
+# Loom bug report — default settings
+vidwise https://loom.com/share/abc123def
+```
+
+## Output
+
+vidwise creates a single self-contained directory:
+
+```
+vidwise-abc123-2026-02-26/
+├── video.mp4              # Source video
+├── audio.wav              # Extracted audio (16kHz mono)
+├── transcript.txt         # Plain text transcript
+├── transcript.srt         # Timestamped subtitles
+├── transcript.json        # Full Whisper output with segments
+├── frames/                # Key frames every 2 seconds
+│   ├── frame_0m00s.png
+│   ├── frame_0m02s.png
+│   ├── frame_0m04s.png
+│   └── ...
+└── guide.md               # Visual guide with embedded frames (if AI enabled)
+```
+
+The `guide.md` uses relative image paths — open it in any markdown viewer (VS Code, GitHub, Obsidian) and the images render inline.
+
+## How It Works
+
+```
+┌─────────────┐
+│  Video URL   │──→ yt-dlp download
+│  or local    │
+└──────┬───────┘
+       │
+       ▼
+┌──────────────┐     ┌──────────────────┐
+│   ffmpeg     │────→│  audio.wav       │──→ Whisper ──→ transcript.*
+│  (parallel)  │     │  (16kHz mono)    │
+│              │────→│  frames/         │──→ Key frame selection
+│              │     │  (every 2 sec)   │    (pixel diff filtering)
+└──────────────┘     └──────────────────┘
+                              │
+                              ▼
+                     ┌──────────────────┐
+                     │  AI Analysis     │  Claude API, OpenAI API,
+                     │  (optional)      │  or Claude Code (free)
+                     └────────┬─────────┘
+                              │
+                              ▼
+                     ┌──────────────────┐
+                     │   guide.md       │  Structured markdown with
+                     │                  │  embedded frame images
+                     └──────────────────┘
+```
+
+**Smart frame selection:** Not every frame matters. vidwise compares consecutive frames using pixel-difference analysis and only keeps frames where the visual content actually changed. A 10-minute video might have 300 raw frames but only ~40 meaningful ones.
+
+## Claude Code Plugin
+
+If you use [Claude Code](https://docs.anthropic.com/en/docs/claude-code), install vidwise as a plugin for **AI-powered guide generation without needing an API key** — Claude Code's native multimodal AI handles the analysis:
+
+```bash
+# Add the vidwise marketplace and install the plugin
+/plugin marketplace add jpdjere/vidwise
+/plugin install vidwise@vidwise
+
+# Then use it:
+/vidwise:vidwise recording.mp4
+/vidwise:vidwise https://loom.com/share/abc123
+```
+
+For local development or testing, you can also load directly:
+
+```bash
+claude --plugin-dir /path/to/vidwise/plugin
+```
+
+The plugin runs `vidwise --no-guide` for extraction, then uses Claude Code's built-in vision capabilities to analyze frames in parallel — completely free, no API key needed.
+
+## Whisper Model Sizes
+
+| Model | Speed | Quality | Best For |
+|-------|-------|---------|----------|
+| `tiny` | ~1 min/min | Basic | Quick tests, long videos |
+| `base` | ~2 min/min | Good | Short videos |
+| `small` | ~4 min/min | Better | Videos >30 min |
+| `medium` | ~8 min/min | Recommended | Default for most content |
+| `large` | ~16 min/min | Best | When accuracy is critical |
+
+*Speed estimates on Apple M-series. First run downloads model weights (one-time).*
+
+## Contributing
+
+Contributions are welcome! Please open an issue first to discuss what you'd like to change.
+
+```bash
+# Development setup
+git clone https://github.com/jpdjere/vidwise
+cd vidwise
+python -m venv .venv
+source .venv/bin/activate
+pip install -e ".[dev]"
+
+# Run tests
+pytest
+
+# Lint
+ruff check src/
+```
+
+## License
+
+[MIT](LICENSE)

vidwise-0.1.0/assets/banner.svg

@@ -0,0 +1,63 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 900 320" fill="none">
+  <defs>
+    <linearGradient id="g1" x1="0%" y1="0%" x2="100%" y2="100%">
+      <stop offset="0%" style="stop-color:#6366f1" />
+      <stop offset="100%" style="stop-color:#8b5cf6" />
+    </linearGradient>
+    <linearGradient id="g2" x1="0%" y1="0%" x2="100%" y2="100%">
+      <stop offset="0%" style="stop-color:#8b5cf6" />
+      <stop offset="100%" style="stop-color:#a78bfa" />
+    </linearGradient>
+    <linearGradient id="g3" x1="0%" y1="0%" x2="100%" y2="0%">
+      <stop offset="0%" style="stop-color:#6366f1" />
+      <stop offset="50%" style="stop-color:#8b5cf6" />
+      <stop offset="100%" style="stop-color:#a78bfa" />
+    </linearGradient>
+  </defs>
+
+  <!-- Centered icon group -->
+  <g transform="translate(290, 30)">
+    <!-- Video frame -->
+    <rect x="0" y="0" width="140" height="140" rx="20" fill="url(#g1)" />
+
+    <!-- Film strip notches -->
+    <rect x="8" y="8" width="12" height="8" rx="2" fill="white" opacity="0.3" />
+    <rect x="28" y="8" width="12" height="8" rx="2" fill="white" opacity="0.3" />
+    <rect x="48" y="8" width="12" height="8" rx="2" fill="white" opacity="0.3" />
+    <rect x="8" y="124" width="12" height="8" rx="2" fill="white" opacity="0.3" />
+    <rect x="28" y="124" width="12" height="8" rx="2" fill="white" opacity="0.3" />
+    <rect x="48" y="124" width="12" height="8" rx="2" fill="white" opacity="0.3" />
+
+    <!-- Play triangle -->
+    <polygon points="52,38 100,70 52,102" fill="white" opacity="0.95" />
+
+    <!-- Eye overlapping the video frame -->
+    <g transform="translate(85, 32)">
+      <path d="M0,38 Q55,-15 110,38 Q55,91 0,38 Z" fill="url(#g2)" />
+      <circle cx="55" cy="38" r="20" fill="white" />
+      <circle cx="55" cy="38" r="10" fill="#4f46e5" />
+      <circle cx="51" cy="33" r="3.5" fill="white" />
+    </g>
+
+    <!-- Subtle sparkle / AI indicator -->
+    <g transform="translate(178, 12)" fill="#a78bfa">
+      <path d="M8,0 L10,6 L16,8 L10,10 L8,16 L6,10 L0,8 L6,6 Z" opacity="0.7" />
+    </g>
+    <g transform="translate(200, 28)" fill="#c4b5fd">
+      <path d="M5,0 L6.2,3.8 L10,5 L6.2,6.2 L5,10 L3.8,6.2 L0,5 L3.8,3.8 Z" opacity="0.5" />
+    </g>
+  </g>
+
+  <!-- Title -->
+  <text x="450" y="230" text-anchor="middle" font-family="system-ui, -apple-system, 'Segoe UI', Roboto, sans-serif" font-size="72" font-weight="800" fill="#1e1b4b" letter-spacing="-2">
+    vid<tspan fill="url(#g1)">wise</tspan>
+  </text>
+
+  <!-- Tagline -->
+  <text x="450" y="270" text-anchor="middle" font-family="system-ui, -apple-system, 'Segoe UI', Roboto, sans-serif" font-size="21" fill="#6b7280" font-weight="500" letter-spacing="0.3">
+    LLMs can't watch videos. vidwise gives them eyes.
+  </text>
+
+  <!-- Accent line -->
+  <rect x="350" y="285" width="200" height="3" rx="1.5" fill="url(#g3)" opacity="0.4" />
+</svg>