Workflow Builder User Guide¶
Overview¶
The Workflow Builder enables users to create custom workflows for repetitive processes (music videos, slide decks, social media content, reports). Build workflows visually with drag-drop interface, execute steps in parallel for 3-5x speedup, and monitor progress in real-time.
Quick Start¶
1. Create Workflow¶
Navigate to: Dashboard → Workflows → Create Workflow
Options: - From Template: Start with pre-built workflow (music video, slide deck, email campaign) - From Scratch: Build custom workflow step-by-step - From Intent: Describe process in natural language, AI composes workflow
2. Add Steps¶
Workflow Canvas: Drag-and-drop interface
Step Types: - Generate Images: AI image generation (DALL-E, Midjourney, Stable Diffusion) - Generate Video: AI video generation (Runway Gen-3, Pika) - Generate Audio: Music, voiceovers (Eleven Labs, Suno) - Generate Text: Copy, scripts (GPT-4o, Claude) - Composite Media: Combine images, video, audio (FFmpeg) - API Call: HTTP requests to external services - Transform Data: JSON, CSV manipulation - Conditional: If/else branching based on step results - Approval Gate: Pause for human approval
Adding Steps: 1. Click "+ Add Step" button 2. Select step type from library 3. Configure step (see configuration section below) 4. Click "Add to Workflow"
3. Configure Dependencies¶
Types: - Sequential: Step B waits for Step A to complete - Parallel: Steps A and B run simultaneously (3-5x faster) - Conditional: Step C only runs if Step A succeeded
Setting Dependencies: 1. Click on step card 2. Go to "Dependencies" tab 3. Select "Depends On" dropdown 4. Choose previous steps that must complete first 5. Save
Parallel Groups: - Steps with same parallel group number run concurrently - Example: Generate images (group 1) + Generate video (group 1) = both run at same time - Sequential steps: Use different group numbers (0, 2, 3, ...)
4. Execute Workflow¶
Steps: 1. Review workflow configuration 2. Click "Execute Workflow" button 3. Monitor progress in real-time
Execution Features: - Real-time progress bar (0-100%) - Step-by-step status (pending, running, completed, failed) - Parallel execution visualization (multiple steps active) - Cost tracking ($0.00 → estimated total) - Time remaining estimate
Workflow Templates¶
Music Video Generation¶
Duration: ~10 minutes Cost: ~$15-$25 Steps: 1. Analyze audio file (extract tempo, beats, mood) 2. Generate scene descriptions based on audio analysis 3. Parallel: Generate visuals for each scene (10 scenes) 4. Composite video with transitions and effects 5. Render final video with audio sync
Parallelization Benefit: 10 scenes generated simultaneously → 5x faster than sequential
Slide Deck Generation¶
Duration: ~5 minutes Cost: ~$5-$10 Steps: 1. Analyze presentation topic and key points 2. Generate outline (intro, main points, conclusion) 3. Parallel: Generate slide content + images for each slide 4. Apply brand styling (colors, fonts, logos) 5. Generate PDF and PowerPoint files
Email Campaign Generation¶
Duration: ~3 minutes Cost: ~$2-$5 Steps: 1. Analyze target audience and campaign goal 2. Generate subject lines (10 variations) 3. Generate email body (HTML + plain text) 4. Generate images for email (hero, product shots) 5. Generate A/B test variations
Social Media Content Batch¶
Duration: ~8 minutes Cost: ~$10-$15 Steps: 1. Analyze content theme and target platforms 2. Parallel: Generate content for each platform: - Instagram post + caption + hashtags - Twitter thread (5 tweets) - LinkedIn article - Facebook post 3. Generate images for each platform (optimized sizes) 4. Schedule posts (integration with social media tools)
Step Configuration¶
Generate Images Step¶
Configuration:
{
"prompt": "Professional office workspace, modern design",
"count": 5,
"size": "1024x1024",
"style": "photorealistic",
"model": "dall-e-3" // or "midjourney", "stable-diffusion"
}
Output: Array of image URLs
Generate Video Step¶
Configuration:
{
"prompt": "Time-lapse of city skyline at sunset",
"duration_seconds": 30,
"resolution": "1920x1080",
"fps": 30,
"model": "runway-gen3" // or "pika"
}
Output: Video URL
Conditional Step¶
Configuration:
{
"condition": "{{step_3.output.success}} == true && {{step_3.output.quality_score}} > 8.0",
"if_true_step": "deploy_production",
"if_false_step": "regenerate_content"
}
Supported Operators:
- Comparison: ==, !=, >, <, >=, <=
- Logical: &&, ||, !
- Parentheses: (, )
Approval Gate Step¶
Configuration:
{
"title": "Review Generated Content",
"description": "Please review the generated video before deploying to social media",
"approval_timeout_hours": 24,
"required_approvers": 1
}
How it works: 1. Workflow pauses at this step 2. Notification sent to approvers 3. Approvers review content and click "Approve" or "Reject" 4. If approved: Workflow continues 5. If rejected: Workflow stops or branches to "rejection" path 6. If timeout: Workflow fails after 24 hours
Parallel Execution¶
How It Works¶
Sequential Execution (traditional):
Parallel Execution (optimized):
Step 1 (30s) → [Step 2 (60s) + Step 3 (60s) in parallel] → Step 4 (45s) = 135 seconds
Speedup: 31% faster, same cost
Setting Parallel Groups¶
Method 1: Automatic Optimization 1. Build workflow sequentially 2. Click "Optimize Workflow" button 3. System analyzes dependencies 4. Automatically assigns parallel groups 5. Review suggestions, apply
Method 2: Manual Assignment 1. Click on step card 2. Set "Parallel Group" number 3. Steps with same number run in parallel 4. Group 0 = sequential (one at a time) 5. Group 1+ = parallel (concurrent with other steps in same group)
Example:
Step 1: Analyze audio (group 0) - runs first
Step 2: Generate scene 1 (group 1) - runs in parallel with steps 3-5
Step 3: Generate scene 2 (group 1)
Step 4: Generate scene 3 (group 1)
Step 5: Generate scene 4 (group 1)
Step 6: Composite video (group 2) - waits for all group 1 steps
Best Practices¶
✅ Good for Parallel Execution: - Image generation (multiple images) - Video scene generation (multiple scenes) - API calls to different services - Text generation (multiple variations) - Independent data transformations
❌ NOT Good for Parallel Execution: - Steps with dependencies (B needs output from A) - Database transactions (race conditions) - File operations on same file - Sequential processes (step 2 builds on step 1)
Workflow Optimization¶
Cost Optimization¶
Goal: Minimize AI API costs while maintaining quality
Strategies: 1. Model Selection: Use cheaper models for simple tasks - GPT-4o Mini for basic text generation (5x cheaper than GPT-4o) - Stable Diffusion for simple images (10x cheaper than DALL-E 3) 2. Batch Operations: Generate multiple outputs in one API call - Generate 10 images in one request instead of 10 separate requests 3. Caching: Reuse results from previous executions - If input is identical, use cached result (0 cost) 4. Early Termination: Stop workflow if quality threshold not met - Conditional step: If quality < 7.0, stop and regenerate
Cost Savings: 20-30% reduction typical
Time Optimization¶
Goal: Minimize workflow duration
Strategies: 1. Maximize Parallelization: Identify all independent steps, run concurrently 2. Fast Models: Use faster AI models when quality difference < 5% - GPT-4o Turbo: 2x faster than GPT-4o, similar quality 3. Reduce Retries: Set max_retries=1 for non-critical steps 4. Skip Unnecessary Steps: Use conditionals to skip steps when possible
Time Savings: 30-50% reduction typical
Balanced Optimization¶
Goal: Best cost/time ratio
MAGIEVA Default: Automatically selects balanced optimization
How it works: 1. Identify expensive operations (> $1.00 per step) 2. Parallelize expensive operations only 3. Keep cheap operations sequential (< $0.10 per step) 4. Use mid-tier models (good quality, reasonable cost/time)
Monitoring & Debugging¶
Real-Time Monitoring¶
Workflow Execution Dashboard: - Overall progress (0-100%) - Active steps (currently running) - Completed steps (with duration and cost) - Failed steps (with error messages) - Pending steps (waiting for dependencies)
Step Details: - Click on step card to view: - Input parameters - Output data - Logs - Retry history - Cost breakdown
Debugging Failed Workflows¶
Common Failure Reasons:
- API Rate Limit Exceeded
- Solution: Add delay between API calls, reduce parallel steps
-
Retry: Automatically retries after 2s, 4s, 8s (exponential backoff)
-
Model Timeout
- Solution: Reduce complexity (shorter prompts, lower resolution)
-
Retry: Switch to faster model automatically
-
Invalid Output Format
- Solution: Fix conditional expression, update expected output schema
-
Manual Fix: Edit step configuration, retry step
-
Budget Exceeded
- Solution: Increase budget limit or wait until next month
- No Retry: Workflow stops immediately
Retry Options: - Automatic: Failed steps retry 3 times (exponential backoff) - Manual: Click "Retry Step" button in workflow monitor - Skip: Click "Skip Step" to continue workflow (step marked as skipped)
Advanced Features¶
Dynamic Workflow Composition¶
From Natural Language Intent:
Example:
Intent: "Create a social media campaign for product launch. Generate 10 posts for Instagram, 5 tweets, and 1 LinkedIn article. Use images from our asset library."
System generates:
- Step 1: Analyze product and target audience
- Step 2-11: Generate Instagram posts (parallel, group 1)
- Step 12-16: Generate tweets (parallel, group 2)
- Step 17: Generate LinkedIn article (group 3)
- Step 18-28: Generate images for posts (parallel, group 4)
- Step 29: Package all content for review
Benefits: - No manual workflow building - Optimized parallel execution automatically - Adapts to intent complexity
Workflow Versioning¶
How it works: - Each workflow has version number (v1, v2, v3) - Modifying workflow creates new version - Existing executions continue with original version - New executions use latest version
Rollback: 1. Go to workflow settings 2. Click "Version History" 3. Select previous version 4. Click "Restore Version"
Workflow Templates (Marketplace)¶
Create Template from Workflow: 1. Build and test workflow 2. Click "Save as Template" 3. Enter template metadata (name, description, category) 4. Publish to marketplace (private or public) 5. Others can instantiate your workflow template
Troubleshooting¶
Workflow won't start¶
Check: - All required fields filled in - No circular dependencies (A depends on B, B depends on A) - Budget sufficient for estimated cost
Parallel steps executing sequentially¶
Check:
- Steps have same parallel group number
- MAX_PARALLEL_STEPS setting (should be > 1)
- Dependencies: If step has "depends_on", it can't run in parallel with dependency
Workflow stuck at checkpoint¶
Solutions: - Check if approval required (approval gate step) - Click "Resume" button - If checkpoint expired (> 24 hours), restart workflow
Best Practices¶
Design Phase¶
- Start Simple: Build with 3-5 steps, test, then expand
- Test with Sample Data: Use placeholder inputs before real data
- Plan Dependencies: Draw workflow diagram on paper first
- Estimate Costs: Review estimated cost per step before executing
Execution Phase¶
- Monitor Actively: Watch first few executions closely
- Review Outputs: Check step outputs for quality
- Set Approval Gates: For expensive steps (> $5), add approval gate
Optimization Phase¶
- Analyze Metrics: Review cost/time per step after 5+ executions
- Identify Bottlenecks: Which steps take longest? Cost most?
- Apply Optimizations: Parallelize, use cheaper models, add conditionals
- A/B Test: Run both versions, compare results
FAQs¶
Q: What's the maximum number of parallel steps? A: Default: 10 parallel steps. Contact support to increase limit.
Q: Can I schedule workflows to run automatically? A: Yes! Use workflow scheduling feature (cron-like syntax).
Q: How do I share workflows with my team? A: Workflows are organization-level. All team members can view, edit, execute.
Q: Can I integrate external APIs? A: Yes! Use "API Call" step type, configure endpoint, headers, authentication.
Q: What happens if I close my browser during execution? A: Workflow continues in background. You can return later to view results.
Q: Can I export workflow definition? A: Yes! Click "Export" button, download JSON file. Import to another org if needed.
Getting Help¶
- Support: support@magieva.com
- Community: Discord
Last Updated: 2026-02-07 Related Guides: Template Marketplace Guide, Application Generator Guide