AgentSkillsCN

brand-visual-pipeline

统筹完整的视觉资产流程，从品牌文档到成品输出。依次串联视觉资产提示生成器、视觉资产生成器与视觉资产整合器。适用于为品牌生成并整合视觉资产时使用。

SKILL.md
--- frontmatter
name: brand-visual-pipeline
description: Orchestrates the complete visual asset pipeline from brand docs to production-ready output. Chains visual-asset-prompt-generator, visual-asset-generator, and visual-asset-integrator in sequence. Use when generating and integrating visual assets for a brand.

Brand Visual Pipeline

Meta-skill that orchestrates the complete visual asset generation workflow, from reading brand documentation to producing a finished landing page, pitch deck, or website with AI-generated visuals.

Pipeline inspired by: Dilum Sanjaya's workflow (Midjourney -> Nano Banana -> Hunyuan3D -> Gemini Pro), adapted to use brand docs as input instead of manual image searches.

When to Use

  • •User says "generate visual assets for [brand]"
  • •User says "create a landing page with AI visuals for [brand]"
  • •User says "build pitch deck visuals from brand docs"
  • •User says "run the visual pipeline"
  • •Any request combining brand docs + AI image/3D generation + output target

Input Variables

  • •[BRAND_NAME] — Brand name (required)
  • •[OUTPUT_CONTEXT] — Target: landing-page | pitch-deck | website | all
  • •[ASSET_TYPES] — What to generate: hero-image | product-shot | 3d-model | all
  • •[PRODUCT_FOCUS] — Optional: specific product/feature

Required Environment

bash
FAL_KEY=xxx       # fal.ai API key
HF_TOKEN=hf_xxx  # HuggingFace Pro token (optional, for HF fallback)

The Protocol

Step 1: Invoke visual-asset-prompt-generator

Read brand docs and generate structured prompts.

Input: Brand docs directory, output context, asset types Output: brands/{BRAND_NAME}/visual-prompts.json

Verification: Confirm prompts JSON exists and contains entries for requested asset types. Verify brand colors and mood keywords are present in prompts.

Step 2: Invoke visual-asset-generator

Execute AI generation using the prompts.

Input: Prompts JSON from Step 1 Output: assets/{BRAND_NAME}/manifest.json + generated files

For images (fal.ai):

  • •Call Flux 1.1 Pro or Nano Banana via fal.ai API
  • •Download results, create WebP + thumbnails
  • •Typically completes in 10-30 seconds per image

For 3D models (Hunyuan3D):

  • •If Blender MCP connected: use Blender tools
  • •If not: use HuggingFace API or skip with warning
  • •Async polling required — can take 5-10 minutes
  • •Export GLB + viewport render

Verification: Confirm manifest.json exists with status "completed" for all assets. Check files exist on disk.

Step 3: Invoke visual-asset-integrator

Place generated assets into the target output.

Input: Asset manifest + brand visual identity Output: Production-ready HTML/markdown/package

Verification: Open generated HTML in browser to visually verify brand alignment.

Step 4: Report

Generate summary of what was created:

  • •Number of assets generated
  • •Services used (fal.ai, Hunyuan3D, PolyHaven)
  • •Output files with paths
  • •Any errors or skipped items

Example Invocation

code
User: "Generate visual assets for NOESIS and create a landing page"

Pipeline execution:
1. visual-asset-prompt-generator
   - Reads brands/noesis-brand-docs/ (13 docs)
   - Extracts: Deep Ink #1A1A2E, Bone #F5F0E8, Aged Gold #B8860B
   - Mood: grounded, substantial, architectural
   - Avoid: ethereal, mystical, soft
   - Generates: 2 image prompts (hero, product) + 1 3D prompt
   - Output: brands/noesis/visual-prompts.json

2. visual-asset-generator
   - Calls fal.ai Flux 1.1 Pro for hero image (1920x1080)
   - Calls fal.ai Flux 1.1 Pro for product shot (1200x800)
   - Calls Hunyuan3D for product 3D model (polls until done)
   - Post-processes: PNG, WebP, thumbnails
   - Output: assets/noesis/manifest.json

3. visual-asset-integrator
   - Reads manifest + visual identity
   - Generates: noesis-landing-page.html
   - Includes: hero section, product gallery, 3D viewer
   - Brand-aligned CSS: Deep Ink bg, Bone text, Gold CTAs
   - Output: brands/noesis/landing-page/index.html

4. Report
   - 3 assets generated (2 images, 1 3D model)
   - Services: fal.ai (2), Hunyuan3D (1)
   - Output: landing page HTML with all assets integrated

Error Recovery

FailureRecovery
fal.ai unavailableSkip image gen, report which prompts weren't executed
Blender not connectedSkip 3D gen or use HF API fallback
Partial generationContinue with available assets, note gaps in report
Brand docs missingReport which docs are needed, suggest running upstream skills

Integration

  • •Chains: visual-asset-prompt-generator -> visual-asset-generator -> visual-asset-integrator
  • •Upstream: visual-identity-core (brand must have visual identity docs)
  • •Integrates with: pitchdeck-skill (deferred — Phase 4), campaign-page-builder
  • •External APIs: fal.ai, HuggingFace, Blender MCP