shellbot-creative

Create high-end creative outputs by chaining providers intentionally:

1. 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: INLINECODE14

Authentication and provider checks

Support any of these providers:

• - FREEPIK_API_KEY for Freepik API
• INLINECODE16 for fal.ai API
• INLINECODE17 (or infsh login) for Nano Banana 2 via inference.sh

Before any generation call:

CODEBLOCK0

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.

CODEBLOCK1

Then create a provider/model routing recommendation:

CODEBLOCK2

$0 = asset

Generate still assets for scenes.

• - Nano Banana 2 default (consistency + editing):

• - Freepik high-fidelity product image (Mystic):

• - fal fast concepting (Flux):

$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: INLINECODE25
• fal default when Freepik key is unavailable: INLINECODE26

Always save scene outputs in a structured project folder:

CODEBLOCK6

$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 INLINECODE29
• Target: deliver 10-240s cue matching storyboard duration

$0 = remotion

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

CODEBLOCK7

Generate a Remotion manifest from the storyboard:

CODEBLOCK8

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

$0 = pipeline

Run end-to-end sequence:

1. 1. plan from brief
2. INLINECODE34 generation per scene
3. INLINECODE35 and upscale selected assets
4. INLINECODE36 generation for hero scenes
5. INLINECODE37 + INLINECODE38
6. INLINECODE39 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
• INLINECODE44 - requirement -> provider/model route
• INLINECODE45 - storyboard -> Remotion-ready manifest
• INLINECODE46 - full local dry run that outputs storyboard, routes, manifest, and runnable API plan
• INLINECODE47 - build a distributable .tar.gz of the skill
• INLINECODE49 - install the skill folder into an OpenClaw skills directory
• INLINECODE50 - durable standards for creative quality
• INLINECODE51 - provider/model decision matrix
• INLINECODE52 - end-to-end recipes for common outcomes
• INLINECODE53 - concrete command samples
• INLINECODE54 - Remotion-specific finishing guidance
• INLINECODE55 - starter Remotion composition
• INLINECODE56 - 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