Claude Skill

video-generation

Generate short videos with the generate_video tool (Veo, Sora, Hailuo) - scripts, storyboards, scene-by-scene AI generation, and ffmpeg assembly. Use for social clips, product demos, and ads.

LLM Mart · 0 points · 2 views 0 listing impressions 0 install-command copies
Download Navinspire-ia-navin-navin_skills_video-generation-772383e.zip · 2 KB

Install

skills CLI npx skills add https://github.com/Navinspire-ia/navin/tree/main/navin/skills/video-generation
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install navinspire-ia-navin@llmmart
Git git clone https://github.com/Navinspire-ia/navin.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole navinspire-ia/navin collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Video Generation

Overview

Four layers of video work: the script (always), AI generation with generate_video, image-to-video from product shots, and assembly (montage).

Tooling

  • generate_video - text-to-video or image-to-video through the configured provider (Google Veo via Gemini, OpenAI Sora, MiniMax Hailuo). Returns a persistent artifact (vid_...) with a local path. Generation takes minutes; tell the user before launching several clips.
  • generate_image - produce a key visual or first frame, then animate it by passing the artifact path as reference_image to generate_video.
  • If generate_video is not in the tool list, video generation is not enabled: deliver script + storyboard + ready-to-use prompts instead, and point the user to Settings → Video to enable a provider.

1. Script & storyboard (the real value)

30s vertical clip structure:

  • 0-3s hook - question, striking number, or surprising visual (decides everything)
  • 3-20s body - one point, illustrated
  • 20-30s payoff + CTA

Storyboard format: | Sec | Visual | VO/text overlay | generate_video prompt | - validate before generating anything.

2. AI generation with generate_video

Generate scene by scene from the storyboard. Each prompt needs:

  • Subject and action ("a barista pours latte art in slow motion")
  • Camera movement ("slow dolly-in", "orbit", "static tripod shot")
  • Style and lighting ("warm morning light, shallow depth of field, cinematic")
  • On-screen text quoted exactly, if any
generate_video(
  prompt="Close-up of a matte-black wireless earbud rotating on a marble pedestal, studio lighting, soft reflections, slow 360 orbit, premium product commercial style",
  aspect_ratio="9:16",
  duration_seconds=8
)

Image-to-video for product consistency: generate or reuse a packshot first (see product-visuals), then animate it:

generate_video(
  prompt="Animate this product shot: gentle camera push-in while soft particles of steam rise, keep the product and label unchanged",
  reference_image="/path/to/media/generated/2026-07-21/img_ab12cd34ef56.png",
  aspect_ratio="16:9"
)

Aspect ratios per platform: 9:16 TikTok/Reels/Shorts, 16:9 YouTube/web, 1:1 feed.

3. Assembly (the montage tool, never raw ffmpeg)

Do not hand-write ffmpeg commands through exec: the montage tool already does the editing with progress, cancel, probing and platform canvases (see montage-studio).

  • montage(action=probe, path=...) - real clip length before trimming or placing transitions.
  • montage(action=assemble, visuals=..., durations=..., trims=..., transition=fade, music=..., voice=..., srt=..., profile=...) - one MP4 from clips, stills, music, narration and burned subtitles.
  • montage(action=timeline_save, ...) when the user may adjust the cut in Montage Studio, then timeline_render.
  • montage(action=package, path=..., profiles=default) - 16:9 / 9:16 / 1:1 platform exports.
  • For a product demo, promo or launch film, follow the "Premium demo video" recipe in montage-studio and the motion-design playbook (storyboard, scripted capture, polish, motion scenes, premium assemble, video QA loop). AI clips from generate_video are B-roll inside that edit, not the deliverable on their own.

Subtitles: always (most social video plays muted) - generate the .srt from the script.

Workflow

  1. Clarify: platform (ratio! duration!), goal, brand constraints.
  2. Script + storyboard with one generate_video prompt per scene → user validation.
  3. Generate clips scene by scene; check each artifact before the next.
  4. Assemble with montage(action=assemble) or a timeline; subtitle; package per-platform exports with montage(action=package); deliver via the message tool with artifact paths in media.

Rules

  • Hook first: write 5 hooks, keep 1 (copywriting-agent craft).
  • Brand voice and visual tokens apply (brand-voice-manager, template-manager).
  • Rights: only use assets (music, footage) the user owns or that are licensed.
  • Keep raw artifact paths internal; deliver via the message tool media parameter.
  • For full ad campaigns (concepts, declinations, copy + visuals + video), use ad-creative-generator.
Files (navin)
  • SKILL.md 4.4 KB
    ---
    name: video-generation
    description: Generate short videos with the generate_video tool (Veo, Sora, Hailuo) - scripts, storyboards, scene-by-scene AI generation, and assembly with the montage tool. Use for social clips, product demos, and ads.
    metadata: {"navin":{"emoji":"🎥","category":"marketing"}}
    ---
    
    # Video Generation
    
    ## Overview
    
    Four layers of video work: the script (always), AI generation with `generate_video`, image-to-video from product shots, and assembly (`montage`).
    
    ## Tooling
    
    - **`generate_video`** - text-to-video or image-to-video through the configured provider (Google Veo via Gemini, OpenAI Sora, MiniMax Hailuo). Returns a persistent artifact (`vid_...`) with a local path. Generation takes minutes; tell the user before launching several clips.
    - **`generate_image`** - produce a key visual or first frame, then animate it by passing the artifact path as `reference_image` to `generate_video`.
    - If `generate_video` is not in the tool list, video generation is not enabled: deliver script + storyboard + ready-to-use prompts instead, and point the user to Settings → Video to enable a provider.
    
    ## 1. Script & storyboard (the real value)
    
    30s vertical clip structure:
    - **0-3s hook** - question, striking number, or surprising visual (decides everything)
    - **3-20s body** - one point, illustrated
    - **20-30s payoff + CTA**
    
    Storyboard format: `| Sec | Visual | VO/text overlay | generate_video prompt |` - validate before generating anything.
    
    ## 2. AI generation with `generate_video`
    
    Generate scene by scene from the storyboard. Each prompt needs:
    
    - Subject and action ("a barista pours latte art in slow motion")
    - Camera movement ("slow dolly-in", "orbit", "static tripod shot")
    - Style and lighting ("warm morning light, shallow depth of field, cinematic")
    - On-screen text quoted exactly, if any
    
    ```text
    generate_video(
      prompt="Close-up of a matte-black wireless earbud rotating on a marble pedestal, studio lighting, soft reflections, slow 360 orbit, premium product commercial style",
      aspect_ratio="9:16",
      duration_seconds=8
    )
    ```
    
    Image-to-video for product consistency: generate or reuse a packshot first (see `product-visuals`), then animate it:
    
    ```text
    generate_video(
      prompt="Animate this product shot: gentle camera push-in while soft particles of steam rise, keep the product and label unchanged",
      reference_image="/path/to/media/generated/2026-07-21/img_ab12cd34ef56.png",
      aspect_ratio="16:9"
    )
    ```
    
    Aspect ratios per platform: 9:16 TikTok/Reels/Shorts, 16:9 YouTube/web, 1:1 feed.
    
    ## 3. Assembly (the `montage` tool, never raw ffmpeg)
    
    Do not hand-write ffmpeg commands through `exec`: the `montage` tool already does the editing with progress, cancel, probing and platform canvases (see `montage-studio`).
    
    - `montage(action=probe, path=...)` - real clip length before trimming or placing transitions.
    - `montage(action=assemble, visuals=..., durations=..., trims=..., transition=fade, music=..., voice=..., srt=..., profile=...)` - one MP4 from clips, stills, music, narration and burned subtitles.
    - `montage(action=timeline_save, ...)` when the user may adjust the cut in Montage Studio, then `timeline_render`.
    - `montage(action=package, path=..., profiles=default)` - 16:9 / 9:16 / 1:1 platform exports.
    - For a product demo, promo or launch film, follow the "Premium demo video" recipe in `montage-studio` and the `motion-design` playbook (storyboard, scripted capture, polish, motion scenes, premium assemble, video QA loop). AI clips from `generate_video` are B-roll inside that edit, not the deliverable on their own.
    
    Subtitles: always (most social video plays muted) - generate the .srt from the script.
    
    ## Workflow
    
    1. Clarify: platform (ratio! duration!), goal, brand constraints.
    2. Script + storyboard with one `generate_video` prompt per scene → user validation.
    3. Generate clips scene by scene; check each artifact before the next.
    4. Assemble with `montage(action=assemble)` or a timeline; subtitle; package per-platform exports with `montage(action=package)`; deliver via the `message` tool with artifact paths in `media`.
    
    ## Rules
    
    - Hook first: write 5 hooks, keep 1 (`copywriting-agent` craft).
    - Brand voice and visual tokens apply (`brand-voice-manager`, `template-manager`).
    - Rights: only use assets (music, footage) the user owns or that are licensed.
    - Keep raw artifact paths internal; deliver via the `message` tool `media` parameter.
    - For full ad campaigns (concepts, declinations, copy + visuals + video), use `ad-creative-generator`.
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.