shellbot-creative（贝壳机器人创意）

shellbot-creative

Create high-end creative outputs by chaining providers intentionally:

1. Generate reusable assets (Nano Banana 2 / Freepik / fal)
2. Generate scene motion (Freepik Kling / fal video models)
3. Add voice and music (Freepik ElevenLabs voiceover + music or fal equivalents)
4. Assemble and polish in Remotion

Use this skill as a production orchestrator, not as isolated single calls.

Arguments

• Command: $0 (plan | asset | edit | video | voice | music | remotion | pipeline | status | sample)
• Arg 1: $1 (provider, model, or workflow type)
• Arg 2+: $2, $3, etc.
• All args: $ARGUMENTS

Authentication and provider checks

Support any of these providers:

• FREEPIK_API_KEY for Freepik API
• FAL_KEY for fal.ai API
• INFERENCE_API_KEY (or infsh login) for Nano Banana 2 via inference.sh

Before any generation call:

[ -n "$FREEPIK_API_KEY" ] && echo "Freepik ready"
[ -n "$FAL_KEY" ] && echo "fal ready"
command -v infsh >/dev/null && echo "infsh available"

If a requested provider is not authenticated, route to another available provider and clearly explain the fallback.

Provider routing (opinionated defaults)

• Asset ideation / strong multi-image consistency: Nano Banana 2
• Photoreal product hero, posters, typography, upscale/edit suite: Freepik
• Fast model exploration or niche fal endpoint needs: fal.ai
• Final timeline composition, transitions, captions, audio mix: Remotion

Read references/provider-matrix.md for exact model choices.

Command behavior

$0 = plan

Turn a user brief into a production storyboard and shot list.

python3 scripts/creative_brief_to_storyboard.py \
  --brief "Launch video for a new fitness app aimed at busy professionals" \
  --format product-marketing \
  --duration 45 \
  --aspect-ratio 16:9 \
  --out storyboard.json

Then create a provider/model routing recommendation:

python3 scripts/creative_provider_router.py \
  --goal full-video \
  --priority quality \
  --needs-consistency true \
  --needs-typography true

$0 = asset

Generate still assets for scenes.

• Nano Banana 2 default (consistency + editing):

infsh app run google/gemini-3-1-flash-image-preview --input '{
  "prompt": "Premium SaaS dashboard on a laptop, soft studio light",
  "aspect_ratio": "16:9",
  "num_images": 4,
  "resolution": "2K"
}'

• Freepik high-fidelity product image (Mystic):

curl -s -X POST "https://api.freepik.com/v1/ai/mystic" \
  -H "x-freepik-api-key: $FREEPIK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt":"Studio product shot of matte black earbuds on reflective surface","resolution":"2k","styling":{"style":"photo"}}'

• fal fast concepting (Flux):

curl -s -X POST "https://queue.fal.run/fal-ai/flux-2" \
  -H "Authorization: Key $FAL_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt":"Cyberpunk storefront, rainy night","image_size":"landscape_16_9"}'

$0 = edit

Perform image editing/upscaling/background work before animation.

• Prefer Freepik edit endpoints for deterministic production flows.
• Prefer Nano Banana 2 when iterative instruction-based edits are needed with multiple references.
• Use Freepik upscalers for final delivery quality.

See references/workflow-recipes.md for edit chains.

$0 = video

Generate motion scenes from prompts or keyframes.

• Freepik default for premium multi-shot video: kling-v3-omni-pro
• fal default when Freepik key is unavailable: fal-ai/kling-video/v2/image-to-video

Always save scene outputs in a structured project folder:

mkdir -p ./creative-output/{assets,scenes,audio,final,manifests}

$0 = voice

Generate narration.

• Default: Freepik ElevenLabs voiceover endpoint
• Fallback: fal text-to-speech model if user requests fal-only flow

Use concise script-first narration lines generated from storyboard scenes.

$0 = music

Generate background music and keep loudness headroom for voiceover.

• Default: Freepik music-generation
• Target: deliver 10-240s cue matching storyboard duration

$0 = remotion

Assemble final video with composited assets, generated clips, captions, and audio mix.

npx create-video@latest creative-video
cd creative-video

Generate a Remotion manifest from the storyboard:

python3 ../scripts/remotion_manifest_from_storyboard.py \
  --storyboard ../storyboard.json \
  --fps 30 \
  --voiceover-url "https://example.com/voice.mp3" \
  --music-url "https://example.com/music.mp3" \
  --out ./src/creative-manifest.json

Then implement composition using assets/remotion/ProductMarketingTemplate.tsx as a baseline.

$0 = pipeline

Run end-to-end sequence:

1. plan from brief
2. asset generation per scene
3. edit and upscale selected assets
4. video generation for hero scenes
5. voice + music
6. remotion assembly + render

If the user asks for a product marketing or explainer video, this is the default command path.

$0 = status

Check async task status for Freepik/fal requests and report:

• current state
• queue position if available
• failed-step diagnosis
• next action

$0 = sample

Load ready-to-run workflows from references/samples.md.

Quality bar for "outstanding" creative output

Always enforce:

• Clear narrative arc: hook -> value -> proof -> CTA
• Visual consistency across scenes (same product, palette, style)
• Voice/music balance (voice intelligibility first)
• Captions and on-screen text optimized for mobile safe zones
• Brand-safe outputs and explicit user approval before final render

Production workflow policy

• Prefer 16:9 for explainers and product demos unless user asks otherwise
• Prefer 9:16 for shorts/reels/tiktok style cuts
• Generate 2-4 variations of key hero scenes, then converge
• Keep source assets reusable (clean backgrounds, layered variants when possible)
• For revisions, edit prompts/scene timing first before regenerating everything

Local resources included in this skill

• scripts/creative_brief_to_storyboard.py - brief -> timed scene plan
• scripts/creative_provider_router.py - requirement -> provider/model route
• scripts/remotion_manifest_from_storyboard.py - storyboard -> Remotion-ready manifest
• scripts/run_full_dry_run.py - full local dry run that outputs storyboard, routes, manifest, and runnable API plan
• scripts/package_skill.sh - build a distributable .tar.gz of the skill
• scripts/install_skill.sh - install the skill folder into an OpenClaw skills directory
• references/creative-guidelines.md - durable standards for creative quality
• references/provider-matrix.md - provider/model decision matrix
• references/workflow-recipes.md - end-to-end recipes for common outcomes
• references/samples.md - concrete command samples
• references/remotion-playbook.md - Remotion-specific finishing guidance
• assets/remotion/ProductMarketingTemplate.tsx - starter Remotion composition
• assets/remotion/Root.tsx - composition registration sample

References

• OpenClaw skills format and frontmatter: https://docs.openclaw.ai/tools/skills
• OpenClaw slash command behavior ($0, $1, $ARGUMENTS): https://docs.openclaw.ai/tools/slash-commands
• OpenClaw metadata config (metadata.openclaw): https://docs.openclaw.ai/tools/skills-config
• OpenClaw creating skills guide: https://docs.openclaw.ai/tools/creating-skills