CNAPS Studio User Guides

The complete guide to building AI-powered image, video, and audio workflows.

CNAPS Studio is a no-code AI workflow builder that lets anyone automate image editing, video production, background removal, upscaling, OCR, and speech-to-text by connecting visual building blocks. This guide covers everything from your first flow to batch processing

Table of Contents

1. What is CNAPS Studio?
2. Understanding the Basics
3. Platform Guide
4. Common Challenges & Solutions
5. API Integration
6. MCP Guide
7. n8n Guide
8. Zapier Guide
9. Glossary
10. Additional Resources

What is CNAPS Studio?

Two ways to use it

Who It's For

What Problems It Solves

Pricing & Plans

Understanding the Basics

New here? Complete the Quickstart Guide first — you'll build your first flow in 5 minutes.

What is a flow?

A flow is a series of connected steps that process your data automatically. Each step is called a node, and nodes are connected by lines that show how data moves.

Think of it like a factory assembly line:

1. Raw materials enter (Input)
2. Each station does one job (AI Models & Tools)
3. Finished product exits (Output)

Node types

Input Nodes (data entry)

• Image Loader — upload images (JPG, PNG, WEBP, AVIF)
• Text Input — enter text data
• Video Loader — upload video files

AI Model Nodes (smart processing)

→ Browse all models: AI Model Hub

Tool Nodes (data manipulation)

• Blur & Effects — Image Blur (Fast / Standard / Simple), Lens Blur, Selective Blur
• Color Adjustments — Grayscale, Inverse
• Comparison — Image Compare, Video Compare, Text Compare
• Image Blending — Image Add, Image Multiply (now with auto-fit), Layer Compose
• Image Masking — Masker by Class / by Coord, Manual Masker, Object Picker
• Image Resizing — Resize, Conditional Resize, Resize to Match
• Selection & Extraction — Crop by Class / Coord, Manual Crop, Image Selector, Image Gate by Text, Color Pickers, Video Frame Extract, List Extract / List Inject
• Text & Annotation — Draw Text, Text Concatenate, Text Decorator
• Video Editing — Video Split / Reassemble, Video Trim, Video List Extract, Remove Dead Zone, Video Reframe (9:16), Video Caption, Video Subtitle (Spoken), Hook Teaser, Video Narrate, Narration Mux

Output Nodes (View Results)

Where you see the final results.

• Image Viewer — display processed images (resizable)
• Text Viewer — display text results and data
• Video Viewer — display generated/processed videos

AI-Generated Content Labels

When your flow uses AI models to generate or manipulate content, CNAPS Studio automatically labels the output — both in the viewer and in downloaded files.

Chain detection: If your flow has FLUX (image gen) → BiRefNet (segmentation) → Output, the segmentation output still shows the "AI-generated · FLUX Schnell" badge because the upstream chain contains generation. The badge tracks the earliest AI generation step in the flow.

Why it's there:

• EU AI Act Article 50 requires AI-generated content to be identifiable
• The badge tells anyone viewing the output that it came from an AI model — useful for compliance, journalism ethics, and end-user trust
• CNAPS also embeds machine-readable provenance metadata in downloaded files

When it appears:

• ✅ Flow includes any generative model (image generation, image editing, video generation, TTS, multimodal LLM, image captioning)
• ✅ Flow includes OCR (labeled as "AI-extracted" not "generated" — to preserve accuracy)
• ❌ Pure utility flows (upscaling, denoising, resize, blur, object detection) — no badge

Connecting Nodes: The Color System

Nodes have colored connection points (called "ports") showing what data type they handle:

• 🔵 Blue dots = Image data
• 🔴 Red dots = Text data
• Light Blue dots = Video data
• 🔵🔴 Both colors = Nodes that handle both types

Understanding Usage Capacity

CNAPS Studio measures your usage by GPU processing time - essentially, how much computing power your flows consume.

• Each plan includes a set amount of usage capacity
• Complex AI models use more capacity than simple tools
• Capacity resets every 5 hours (rolling window)
• Monitor real-time usage: Dashboard → Usage
• If you exceed capacity: processing pauses until the next reset, or upgrade for more

Easy UI — Instant AI, No Workflow Needed 🆕

How it works:

1. Open Easy UI and browse the card gallery (Virtual Try-On, Beyond Upscaling, Identity Redacted, and more)
2. Drop in your text, image, or video
3. Hit Run — your result appears on the same page
4. Find past results anytime via the My Results button

Good to know:

• Works on mobile — browse cards, upload, run, and view results from your phone
• Per-card parameter menus let you fine-tune each run without opening the canvas
• Cards that use commercial APIs show a live API connection status, so you know before you run
• Ready for more? Click "Open CNAPS Studio" on any card to load the same flow onto the full canvas and customize it

Ultimate Privacy Protection — One-Click Anonymization 🆕

A standalone web app for privacy-safe images — no workflow, no setup:

1. Upload an image (PNG, JPEG, WebP, AVIF, BMP — up to 50 MB)
2. Choose what to hide: Face, Tattoo, License Plate, Person, or Custom (type anything, e.g. "company logo")
3. Get your anonymized image back on the same page

Platform Guide

Dashboard Overview

The Dashboard is your control center for creating, managing, and running AI workflows.

→ New user? Complete the Quickstart Guide to familiarize yourself with the interface.

🧩 Creating and Managing Flows

Starting a New Flow

3️⃣ Method 3: Start from Easy UI 🆕

Run any Easy UI card, then click "Open CNAPS Studio" to drop the underlying flow onto the canvas — a great way to learn how flows are built.

Configuring Nodes

Each node has settings that control its behavior. Configuration options appear directly on the node card when you select it.

Click any node to see its settings directly on the node card. Common options:

• Class — which object type to detect or process (e.g., "person", "car")
• Confidence Score — minimum certainty threshold (0–100%); higher = fewer but more accurate results
• Scale Factor — how much to resize or upscale
• Kernel Size — size of processing filter; larger = stronger effect
• Diffusion Steps — refinement iterations for generative models; more = higher quality but slower
• Random Seed — same seed = identical results every time (great for testing)

Running Your Flow

Single Run (testing & development)

1. Upload input data to Input nodes
2. Verify connections
3. Click Run Flow (top right)
4. Watch progress — the currently-processing node glows
5. View results in Output nodes; click to see full detail; download if needed

Visual cues during a run: animated button states (Default → Running → Done/Failed), glowing active node, smooth step-by-step progress bars, and background execution — flows keep running even if you switch tabs.

Batch Run (Production Processing)

Process up to 100 files through the same workflow at once.

Three input modes per node:

How to run a batch:

1. Click Batch Run (top right) — one panel opens per input node (#1, #3, …)
2. Choose Batch / Shared / Flow Value per input
3. Drag & drop files or folders (up to 100 items)
4. Files in the same row are paired and run together — use Export CSV / Import CSV to reorder pairings for complex batches
5. Click Run Batch, then monitor at Dashboard → Batch Runs
6. The completion email now links straight to your results — one click from inbox to download
7. Download everything as a ZIP, or preview and download individual files

(Developers can also trigger batches via API — see API Integration below.)

Sharing Flows

Three access modes plus per-person permissions — precise control, like Google Docs:

People with access: invite collaborators by name or email as Editor (view, run, modify) or Viewer (view & run only). Optional extras: require an access code, and notify invitees by email.

Real-time collaboration: invite Editors and build together — everyone sees each other's cursors and changes live on the canvas, and anyone can Run Flow or Batch Run.

Working with external models

Use AI services from Anthropic, OpenAI, and Google without consuming your CNAPS capacity — they bill through your own provider accounts.

Current external lineup:

• Image/Video: GPT Image 2.0, Nano Banana, Sora2, VEO, Video Analysis (Gemini) — feed a video and get transcripts, timecodes, or highlight summaries
• Language: ChatGPT (now incl. GPT-5.5 / GPT-5.5 Pro), Claude (now incl. Claude Opus 4.7), Gemini (now incl. Gemini 3.5 Flash)

Setup:

1. Get API keys — Anthropic · OpenAI · Google
2. Paste them in Dashboard → Workspace Settings → Integrations
3. In the Component panel, external models show white text when ready (gray = needs API key)

Working with Claude (MCP)

CNAPS Studio integrates with Claude via MCP (Model Context Protocol) — describe what you want in plain language and Claude assembles, runs, and debugs the workflow inside the chat:

• Smart pipeline recommendations — multi-step requests like "colorize and upscale" become the correct pipeline automatically
• Context-aware model selection — e.g., PiSA-SR for quality, SMFANet+ for speed, Swin2SR for large images
• Actionable error diagnosis — plain-language explanations and fixes instead of generic failures
• End-to-end execution in chat — no need to open the Studio and click Run manually

Connecting with n8n

Trigger CNAPS Studio flows from your existing n8n automations — chain image, video, and multimodal AI processing into any n8n workflow using your CNAPS.ai API key:

• No-code trigger — kick off a CNAPS.ai flow from any n8n node (webhook, schedule, email, Slack, etc.)
• Batch and single-run support — send one file or many through the same workflow
• Pass results downstream — return the CNAPS.ai output (image URLs, extracted text, generated video) into the rest of your n8n pipeline

→ Full setup, node reference, and example recipes: see the n8n Guide.

Connecting with Zapier

Run CNAPS Studio flows as steps inside a Zap, alongside the 9,000+ apps Zapier already automates:

• Any trigger, any destination — start from Gmail, Slack, Airtable, Shopify, a form, or a schedule, then send results wherever they need to go
• Predictable, repeatable runs — you pick the tools and their order when you build the Zap, so every run behaves identically
• Sign in with OAuth — connect with your CNAPS.ai account; no API key to paste, store, or rotate

→ Full setup, field mapping, and known pitfalls: see the Zapier Guide.

Common Challenges & Solutions

This section addresses the most frequent issues users encounter, especially those new to AI workflows.

Challenge #1: "I don't understand what to do"

This is the #1 feedback from beginners. You're not alone! Here's a structured approach to overcome this:

First: Complete the Quickstart If you haven't already, go through our Quickstart Guide. Building that first simple flow gives you essential hands-on experience.

Learning strategy on the canvas:

1. Phase 1 — Don't build from scratch yet. Open a sample template, click Run Flow with defaults, and watch what each node produces.
2. Phase 2 — Make small changes. Change just ONE thing (input image, one setting, one class name), run again, compare.
3. Phase 3 — Build your own. Start with 3–4 nodes (Input → One AI Model → Output), get it working, then add complexity gradually.

Common confusion points:

• Which AI model should I use? Start with 🔥 Hot Models, compare 2–3 similar ones, check star ratings, and read the AI Model Hub
• What do these settings mean? Start with defaults; change one at a time; see the Glossary below
• What connects where? Match the colors, data flows left → right, Input → Processing → Output
• Flow not doing what I expected? Add Output nodes after EACH step to see intermediate results

Still stuck? Watch video tutorials or browse CNAPS Community for working examples to fork.

Challenge #2: "The AI model didn't work as I expected"

This is common because AI models have different training and vocabularies.

Solution 1: Try Different Models

CNAPS Studio provides multiple models for similar tasks:

Solution 2: Resize Strategically

Many models work best at specific sizes:

• 512x512: Most computer vision models
• 1024x1024: Higher resolution models
• Original size: OCR, text extraction

If detection fails:

• Try resizing to 512x512 or 1024x1024
• Very large images might lose accuracy
• Very small images might miss detail

Challenge #3: "My flow is taking too long"

Long processing times usually have specific causes with specific solutions.

Common Causes & Fixes:

Cause 1: Processing Very Large Images

Symptoms:

• Single image takes 30+ seconds
• GPU usage maxed out
• Slow progress through nodes

Solution: Resize BEFORE processing:

Input → Resize (512x512) → [AI Models] → Upscale → Output

Why this works:

• Most AI models are optimized for 512x or 1024x
• Smaller inputs process 10x faster
• Upscale at the END to restore quality
• No visible quality loss for most tasks

Cause 2: Chaining Multiple Upscaling Models

Symptoms:

• Flow takes minutes per image
• Multiple upscalers in sequence
• Capacity depletes quickly

Problem flow:

Input → 4x Upscale → 4x Upscale → 2x Upscale = 32x total (extremely slow)

Solution: Only upscale once at the end:

Input → [Process at 512x] → Single 2x Upscale → Output

If you need 4x:

• Use one 4x upscaler, OR
• Use two 2x upscalers maximum
• Never more than 2 upscalers total

Cause 3: Using Too Many Heavy Models

Heavy models:

• Upscalers (PiSA-SR, SMFANet+)
• Generative models (Image Generation, Object Removal)
• Complex segmentation

Solution: Simplify your flow:

• Remove redundant processing steps
• Use lighter models where possible
• Test each model individually first
• Consider if you need ALL those steps

Cause 4: Batch Processing Large Numbers

Symptoms:

• Batch run shows "Processing" for hours
• Hundreds of files in queue

This is actually normal!

Understanding:

• Batch processing is meant for unattended operation
• Large batches naturally take time
• System processes files sequentially
• Progress shown in Batch Runs page

Solutions:

• ✅ Expected: Let it run (check back later)
• ✅ Faster: Break into smaller batches (50-100 files)
• ✅ Overnight: Schedule during off-hours

Optimization Tips:

For Faster Processing:

1. Resize to 512x as first step
2. Use 2x upscaling instead of 4x
3. Minimize number of AI models in chain
4. Test on small batches first
5. Remove unnecessary processing steps

For Quality vs Speed Trade-offs:

• Prioritize speed: Resize smaller, use 2x upscale, fewer models
• Prioritize quality: Larger inputs, 4x upscale, more refinement
• Balance: 512x processing + single 2x upscale works for most

Challenge #4: "I exceeded my usage capacity"

Your capacity resets every 5 hours automatically, but here's what to do:

Immediate Solutions:

Option 1: Wait for Reset

• Check Dashboard → Usage for timer
• Capacity resets every 5 hours (rolling window)
• Plan work around reset schedule

Option 2: Upgrade Your Plan

• Dashboard → Billing
• Higher tiers = more capacity per 5-hour period
• Compare plans to see limits

Optimize Your Flows:

• Resize images to 512x before processing
• Use 2x upscaling instead of 4x when possible
• Remove unnecessary model chaining
• Test with small batches first

Spread Usage Throughout Day:

• Don't process everything at once
• Utilize the 5-hour reset cycle
• Process in smaller batches across resets

Reserve Capacity:

• Save capacity for critical work
• Process non-urgent items during off-peak
• Monitor usage regularly

Track Patterns:

• Dashboard → Usage shows historical data
• Identify which flows consume most capacity
• Optimize high-usage workflows

Challenge #5: "The output looks wrong or unexpected"

Systematic Troubleshooting:

Step 1: Check Each Node Individually

Add temporary Output nodes after each step:

Input → [Output 1] → Model A → [Output 2] → Model B → [Output 3] → Final Output

Process:

1. Run the flow
2. Check Output 1: Is the input correct?
3. Check Output 2: Did Model A work correctly?
4. Check Output 3: Did Model B work correctly?
5. Identify WHERE things go wrong

Step 2: Check Input Quality

Your input affects output quality:

✅ Good inputs:

• Clear, well-lit images
• Reasonable file size and resolution
• Correct format (JPG or PNG)
• Not heavily compressed or corrupted

❌ Bad inputs:

• Extremely dark or bright images
• Tiny thumbnail-sized images
• Heavily compressed or degraded
• Wrong file format

Quick test: Try the flow with a sample image from a template to see if it works correctly.

Step 3: Try Different Models

If one model produces poor results:

1. Swap it for an alternative model (same category)
2. Try models with higher star ratings
3. Compare results between 2-3 options
4. Different models have different strengths

Step 4: Simplify and Rebuild

If output is completely unexpected:

1. Create a new blank flow
2. Add just Input → Problematic Model → Output
3. Test if model works in isolation
4. Gradually add back other nodes
5. Identify which combination causes issues

API Integration

Getting Your API Key

Managing Your API Keys

After creating a key, you can see it listed in the API Key menu with three options:

• Deactivate - Use when you temporarily want to stop using a key without deleting it
• Regenerate - Use when you think your key might be compromised or want to update it
• Delete - Use when you no longer need this key

Web console only. API keys can only be created, viewed, regenerated, and deleted here in the dashboard. Key management is deliberately not exposed as an MCP tool or API endpoint, so an automated agent cannot issue itself a new key or delete one you depend on.

MCP Guide

Connect Claude to CNAPS Studio via MCP and build workflows through natural language — no dashboard, no manual node-wiring. The server exposes 33 tools, identical whether you connect via OAuth or an API key. The full setup guide, tool reference, and example recipes are on the dedicated page below:

n8n Guide

Trigger CNAPS Studio flows from your existing n8n automations. Chain CNAPS image, video, OCR, and multimodal AI models into any n8n workflow using your CNAPS API key — no code required. The full setup guide, node reference, and example recipes are on the dedicated page below:

Zapier Guide

Run CNAPS Studio flows from Zapier and connect them to the 9,000+ apps Zapier already automates. The MCP Client integration exposes all 33 tools as Zap actions over OAuth, so a fixed pipeline runs the same way every time — upload a file, build a flow, return the result. The full setup guide, field-mapping reference, and known pitfalls are on the dedicated page below:

Glossary

Additional Resources

Getting Started:

Release Note

• Video tutorials
• CNAPS Community

Last Updated: Aug 13th, 2026