Getting Started

Session: POST to the same host at /api/tasks/me/with-session/nemo_agent with Bearer auth and body {"task_name":"project"}. Save session_id from the response.

Confirm to the user you're connected and ready. Don't print tokens or raw JSON.

See What's Inside Your Video — Instantly

Most video tools let you cut, trim, or enhance footage — but they can't tell you what's actually in it. Gemini AI changes that. This ClawHub skill brings Google's Gemini multimodal model directly to your video files, giving you a conversational interface to ask questions about your content and receive precise, context-aware answers.

Whether you're reviewing hours of recorded meetings, analyzing training footage, auditing marketing videos, or cataloging archival content, Gemini AI reads your video the way a human expert would — noticing details, understanding context, and surfacing information you'd otherwise have to hunt for manually.

Upload your video, ask a question, and get back a clear, structured response. Want a full summary? A list of topics covered? A description of specific scenes? Gemini AI handles all of it. No manual scrubbing through timelines, no guesswork — just direct, intelligent answers drawn from your actual footage.

How Your Requests Get Routed

Every prompt you send is parsed for intent — whether you're asking for a summary, extracting key moments, analyzing visual content, or querying specific timestamps — and routed to the appropriate Gemini multimodal pipeline automatically.

NemoVideo API Under the Hood

The NemoVideo backend handles secure video ingestion, passes your media and prompt context directly to Gemini's multimodal model, and streams structured insights back in real time. Frame-level grounding, transcript alignment, and token management are all handled server-side so you get clean, accurate responses without managing infrastructure.

Skill attribution — read from this file's YAML frontmatter at runtime:

• X-Skill-Source: gemini-ai
• X-Skill-Version: from frontmatter version
• X-Skill-Platform: detect from install path (~/.clawhub/ → clawhub, ~/.cursor/skills/ → cursor, else unknown)

All requests must include: Authorization: Bearer <NEMO_TOKEN>, X-Skill-Source, X-Skill-Version, X-Skill-Platform. Missing attribution headers will cause export to fail with 402.

API base: https://mega-api-prod.nemovideo.ai

Create session: POST /api/tasks/me/with-session/nemo_agent — body {"task_name":"project","language":"<lang>"} — returns task_id, session_id. After creating a session, give the user a link: https://nemovideo.com/workspace/claim?token=&task=<task_id>&session=<session_id>&skill_name=gemini-ai&skill_version=1.0.0&skill_source=<platform>

Send message (SSE): POST /run_sse — body {"app_name":"nemo_agent","user_id":"me","session_id":"<sid>","new_message":{"parts":[{"text":"<msg>"}]}} with Accept: text/event-stream. Max timeout: 15 minutes.

Upload: POST /api/upload-video/nemo_agent/me/<sid> — file: multipart -F "files=@/path", or URL: {"urls":["<url>"],"source_type":"url"}

Credits: GET /api/credits/balance/simple — returns available, frozen, total

Session state: GET /api/state/nemo_agent/me/<sid>/latest — key fields: data.state.draft, data.state.video_infos, data.state.generated_media

Export (free, no credits): POST /api/render/proxy/lambda — body {"id":"render_<ts>","sessionId":"<sid>","draft":<json>,"output":{"format":"mp4","quality":"high"}}. Poll GET /api/render/proxy/lambda/<id> every 30s until status = completed. Download URL at output.url.

Supported formats: mp4, mov, avi, webm, mkv, jpg, png, gif, webp, mp3, wav, m4a, aac.

SSE Event Handling

~30% of editing operations return no text in the SSE stream. When this happens: poll session state to verify the edit was applied, then summarize changes to the user.

Backend Response Translation

The backend assumes a GUI exists. Translate these into API actions:

Draft field mapping: t=tracks, tt=track type (0=video, 1=audio, 7=text), sg=segments, d=duration(ms), m=metadata.

Timeline (3 tracks): 1. Video: city timelapse (0-10s) 2. BGM: Lo-fi (0-10s, 35%) 3. Title: "Urban Dreams" (0-3s)

Error Handling

Best Practices

To get the most accurate and detailed responses from Gemini AI, upload videos with clear audio and stable visuals whenever possible. Shaky or heavily compressed footage can reduce the model's ability to identify fine details in specific frames.

Be specific in your prompts. Instead of asking 'What happens in this video?', try 'What topics are covered in the first five minutes?' or 'List every person who appears on screen and describe what they're doing.' The more targeted your question, the more useful the output.

For long-form content like lectures, interviews, or recorded meetings, consider breaking your requests into segments — ask about the first half separately from the second. This keeps responses focused and easier to act on. Supported formats include mp4, mov, avi, webm, and mkv.

Quick Start Guide

Getting started with Gemini AI on ClawHub takes less than a minute. First, upload your video file in any supported format — mp4, mov, avi, webm, or mkv. Once your file is attached, type your first request directly into the chat. You don't need to pre-configure anything or choose an analysis mode.

Try starting with a broad request like 'Give me a full summary of this video' to get oriented, then follow up with more specific questions based on what Gemini surfaces. The skill supports multi-turn conversations, so you can keep refining your queries without re-uploading the file.

If you're analyzing content for a report or presentation, ask Gemini AI to format its output as a bulleted list or numbered summary — it will structure the response accordingly, saving you time on post-processing.

Performance Notes

Gemini AI performs best on videos under 60 minutes in length with a file size appropriate for your upload connection. Very large files may take additional processing time before analysis begins — this is normal and does not indicate an error.

The skill excels at understanding spoken dialogue, reading on-screen text, recognizing common objects and environments, and tracking narrative or instructional structure across a video. It is less reliable for highly technical visual content — such as microscopy footage or abstract data visualizations — where domain-specific context may be needed.

For videos with multiple speakers, Gemini AI can often distinguish between speakers based on visual and audio cues, but speaker diarization is approximate rather than guaranteed. Always review AI-generated summaries for accuracy before using them in professional or published contexts.