Setup Local system AI Clipping Tool
Budget / Salary₹600–1,500
TypeFreelance project
LocationRemote
Posted1 hour ago
Python + Gemini API + FFmpeg Developer for 1-Time Local Multimodal AI Auto-Clipping Setup (Google Meet)
Job Description:
I am looking for an experienced Python, Google Gemini API (`google-genai` SDK), and FFmpeg developer for a one-time, complete setup of a local, multimodal AI video auto-clipping pipeline on my computer. The setup and live verification will be completed over a Google Meet session (screen share).
Project Goal:
Build and configure a repeatable local CLI tool where I pass a long-form video file or gaming/podcast VOD URL (e.g., GTA 6 streams or talking-head videos), optionally enter a custom topic prompt, and automatically generate copyright-safe, vertical 9:16 YouTube Shorts/Reels—using Gemini's native multimodal capabilities with minimal API cost and zero paid SaaS subscriptions.
Required Architecture & Features:
1. Video Ingestion & Multimodal Analysis (Gemini File API):
- Support local video files and `yt-dlp` time-slice downloads (YouTube/Twitch/Kick).
- Upload long-form video/audio directly via the Gemini File API to analyze both spoken dialogue and visual cues (scene changes, reactions, emotions, on-screen action/slides) without requiring a separate transcription tool.
2. Token & Cost Optimization:
- Implement Gemini Context Caching so the same uploaded video can be re-queried multiple times (e.g., new topic searches or extra clips) without re-uploading or repaying full input token costs.
- Configure frame-rate sampling / media resolution settings to keep processing fast and API costs low.
3. AI Highlight, Hook & Copyright-Safe Detection (Gemini 2.5 Pro / Flash):
- Prompt Gemini to identify the top 5–10 standalone, high-retention clips (30–60 seconds) with exact `HH:MM:SS` start and end timestamps.
- Include safety filtering (e.g., skipping story cutscenes/spoilers for gaming clips).
4. Structured JSON Outputs (Response Schema):
- Use Gemini Structured Outputs (Pydantic / Response Schema) so every call returns guaranteed valid JSON containing: start/end timestamps, virality score (1–100), on-screen hook text, reasoning, speaker/subject position tags, metadata, and timestamped captions.
5. Topic-Based Custom Clipping (Prompt Search):
- Support an optional CLI flag (e.g., `--topic "police chases"` or `--topic "moments discussing pricing"`) so Gemini extracts topic-specific clips on demand from the cached video.
6. Smart Framing & Speaker Location Hints (Dynamic 9:16 FFmpeg Cropping):
- Use Gemini visual analysis to tag subject/speaker framing (`left`, `center`, `right`, or `split-screen`) per timestamp segment and translate those hints into automated 9:16 (`1080x1920`) FFmpeg crop/layout commands.
7. Timestamped Subtitles, Translation & Auto-Generated Metadata:
- Extract time-aligned captions from Gemini, burn styled bold subtitles (`.ass`/`.srt`) cleanly into the Shorts safe zone via FFmpeg, and support an optional `--lang` flag for multi-language subtitle translation.
- Auto-generate scroll-stopping on-screen text hooks, YouTube Shorts/Reels titles, descriptions, and hashtags saved alongside each clip.
8. Copyright Audio Shield (Optional Toggle):
- Integrate local `Demucs` (toggleable flag like `--strip-music`) to remove copyrighted background/radio music while keeping clean vocals/dialogue for gaming clips.
9. Clean 1-Command CLI & Handoff:
- Wrap everything into a clean virtual environment with `.env` config so I can run commands like:
`python clip.py "VIDEO_URL_OR_PATH" --clips 5 --topic "funny fails" --lang en --strip-music`
- Include a short `README.txt` cheat sheet with all flags and usage examples.
My System: [Enter OS: macOS / Windows 11, RAM, CPU/GPU]
Session Format: One-time live Google Meet session (estimated 2–3 hours; you can pre-write the script beforehand and spend the call installing, tuning, and testing on my machine).
Budget: [Suggested: ₹4,500–₹8,000 / $60–$120 Fixed Price]
Definition of Done (Before Payment Release):
- Clean local setup (`venv`, `.env`, FFmpeg, `yt-dlp`, `google-genai`, `demucs`) with zero missing dependencies.
- Live end-to-end test in a brand-new terminal window processing 1 long-form video via Gemini File API + Context Caching.
- Live test of a second topic-based query against the cached video to verify Context Caching works.
- Successfully outputs ready-to-upload 1080x1920 (`yuv420p`) Shorts with smart framing, synced burned-in captions (plus 1 translated test), on-screen hook text, and metadata files.
To Apply, Please Answer These 3 Questions:
1. Have you built production scripts using the new `google-genai` Python SDK (specifically File API, Context Caching, and Structured Outputs / Response Schema)?
2. How would you handle FFmpeg dynamic cropping (`left`/`center`/`right`/`split-screen`) and ensure accurate caption timing across segment cuts?
3. What is your fixed price and availability to pre-build the script and complete the live Google Meet setup?
Job Description:
I am looking for an experienced Python, Google Gemini API (`google-genai` SDK), and FFmpeg developer for a one-time, complete setup of a local, multimodal AI video auto-clipping pipeline on my computer. The setup and live verification will be completed over a Google Meet session (screen share).
Project Goal:
Build and configure a repeatable local CLI tool where I pass a long-form video file or gaming/podcast VOD URL (e.g., GTA 6 streams or talking-head videos), optionally enter a custom topic prompt, and automatically generate copyright-safe, vertical 9:16 YouTube Shorts/Reels—using Gemini's native multimodal capabilities with minimal API cost and zero paid SaaS subscriptions.
Required Architecture & Features:
1. Video Ingestion & Multimodal Analysis (Gemini File API):
- Support local video files and `yt-dlp` time-slice downloads (YouTube/Twitch/Kick).
- Upload long-form video/audio directly via the Gemini File API to analyze both spoken dialogue and visual cues (scene changes, reactions, emotions, on-screen action/slides) without requiring a separate transcription tool.
2. Token & Cost Optimization:
- Implement Gemini Context Caching so the same uploaded video can be re-queried multiple times (e.g., new topic searches or extra clips) without re-uploading or repaying full input token costs.
- Configure frame-rate sampling / media resolution settings to keep processing fast and API costs low.
3. AI Highlight, Hook & Copyright-Safe Detection (Gemini 2.5 Pro / Flash):
- Prompt Gemini to identify the top 5–10 standalone, high-retention clips (30–60 seconds) with exact `HH:MM:SS` start and end timestamps.
- Include safety filtering (e.g., skipping story cutscenes/spoilers for gaming clips).
4. Structured JSON Outputs (Response Schema):
- Use Gemini Structured Outputs (Pydantic / Response Schema) so every call returns guaranteed valid JSON containing: start/end timestamps, virality score (1–100), on-screen hook text, reasoning, speaker/subject position tags, metadata, and timestamped captions.
5. Topic-Based Custom Clipping (Prompt Search):
- Support an optional CLI flag (e.g., `--topic "police chases"` or `--topic "moments discussing pricing"`) so Gemini extracts topic-specific clips on demand from the cached video.
6. Smart Framing & Speaker Location Hints (Dynamic 9:16 FFmpeg Cropping):
- Use Gemini visual analysis to tag subject/speaker framing (`left`, `center`, `right`, or `split-screen`) per timestamp segment and translate those hints into automated 9:16 (`1080x1920`) FFmpeg crop/layout commands.
7. Timestamped Subtitles, Translation & Auto-Generated Metadata:
- Extract time-aligned captions from Gemini, burn styled bold subtitles (`.ass`/`.srt`) cleanly into the Shorts safe zone via FFmpeg, and support an optional `--lang` flag for multi-language subtitle translation.
- Auto-generate scroll-stopping on-screen text hooks, YouTube Shorts/Reels titles, descriptions, and hashtags saved alongside each clip.
8. Copyright Audio Shield (Optional Toggle):
- Integrate local `Demucs` (toggleable flag like `--strip-music`) to remove copyrighted background/radio music while keeping clean vocals/dialogue for gaming clips.
9. Clean 1-Command CLI & Handoff:
- Wrap everything into a clean virtual environment with `.env` config so I can run commands like:
`python clip.py "VIDEO_URL_OR_PATH" --clips 5 --topic "funny fails" --lang en --strip-music`
- Include a short `README.txt` cheat sheet with all flags and usage examples.
My System: [Enter OS: macOS / Windows 11, RAM, CPU/GPU]
Session Format: One-time live Google Meet session (estimated 2–3 hours; you can pre-write the script beforehand and spend the call installing, tuning, and testing on my machine).
Budget: [Suggested: ₹4,500–₹8,000 / $60–$120 Fixed Price]
Definition of Done (Before Payment Release):
- Clean local setup (`venv`, `.env`, FFmpeg, `yt-dlp`, `google-genai`, `demucs`) with zero missing dependencies.
- Live end-to-end test in a brand-new terminal window processing 1 long-form video via Gemini File API + Context Caching.
- Live test of a second topic-based query against the cached video to verify Context Caching works.
- Successfully outputs ready-to-upload 1080x1920 (`yuv420p`) Shorts with smart framing, synced burned-in captions (plus 1 translated test), on-screen hook text, and metadata files.
To Apply, Please Answer These 3 Questions:
1. Have you built production scripts using the new `google-genai` Python SDK (specifically File API, Context Caching, and Structured Outputs / Response Schema)?
2. How would you handle FFmpeg dynamic cropping (`left`/`center`/`right`/`split-screen`) and ensure accurate caption timing across segment cuts?
3. What is your fixed price and availability to pre-build the script and complete the live Google Meet setup?
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.