Best Practices

This guide covers tips and patterns for getting the most out of Node Banana.

Workflow Organization

Name Your Nodes

Click node titles to give them meaningful names. Instead of "Prompt" and "Prompt 2", use "Subject Description" and "Style Instructions".

Use Comments

Add comments to complex nodes explaining their purpose. Future you will thank present you.

Group Related Nodes

Create groups for logical sections:

• Input processing
• Generation pipeline
• Output handling

Left-to-Right Flow

Organize workflows so data flows left to right. This matches reading direction and makes workflows easier to understand.

Prompt Writing

Be Specific

Vague prompts get vague results. Include:

• Subject details
• Style and mood
• Lighting conditions
• Camera perspective
• Technical specifications

Instead of:

A cat

Write:

A fluffy orange tabby cat sitting on a windowsill,
golden hour sunlight streaming in, shallow depth of field,
professional pet photography, warm tones

Use Structure

For complex prompts, organize into sections:

Subject: [what you're generating]
Style: [artistic style or technique]
Composition: [framing, angle, layout]
Lighting: [light source, mood, time of day]
Technical: [camera settings, render quality]

Negative Space

Mention what you don't want:

...clean background, no text, no watermarks

Iterate on Prompts

1. Start with a basic prompt
2. Generate and evaluate
3. Add details for what's missing
4. Remove details that cause issues
5. Save your refined prompt

Working with Images

Image Quality Matters

• Use high-resolution source images
• Avoid heavily compressed JPEGs
• Clean, well-lit photos work best

Annotation Best Practices

When annotating images for AI reference:

• Use bright, contrasting colors
• Keep annotations clear and simple
• Circle or highlight, don't obscure
• The AI sees exactly what you draw

Reference Images

When providing reference images to Generate Image nodes:

• Include images that show what you want
• Multiple references can guide style, composition, color
• Conflicting references may confuse the model
• Image-to-image support works across all providers

Execution Strategies

Test Before Batch

Before running a full workflow:

1. Run just the first generation node
2. Check the output
3. Adjust prompts if needed
4. Then run the complete workflow

Use Locked Groups

Lock groups containing completed work. This:

• Prevents accidental re-execution
• Speeds up partial runs
• Preserves successful generations

Run From Node

Use the play button on individual nodes to:

• Re-run specific sections
• Test changes without full execution
• Debug connection issues

Cost Management

Iterate Cheaply

Use faster, cheaper models while iterating. Switch to higher-quality models only for final outputs. Check the Models & Pricing guide for cost comparisons.

Compare Providers

Different providers have different pricing. Replicate and fal.ai often have competitive rates for open-source models.

Check Before Running

Review estimated costs in the header before executing large workflows.

Monitor Cumulative Costs

Track your spending across sessions. Set mental budgets for projects.

File Management

Consistent Naming

Name workflow files descriptively:

• product-shot-workflow.json
• portrait-style-transfer-v2.json

Version Your Workflows

When making significant changes, save as a new version rather than overwriting.

Organize by Project

Create folders for different projects or clients.

Common Patterns

Prompt Enhancement Pipeline

[Your Prompt] → [LLM: Enhance this prompt] → [Generate Image]

Use an LLM to expand simple prompts into detailed generation instructions.

A/B Testing

                    ┌→ [Generate Image: Provider A]
[Prompt] → [Split] ─┤
                    └→ [Generate Image: Provider B]

Compare different providers or models on the same prompt.

Iterative Refinement

[Image] → [Annotate: Mark changes] → [Generate Image: Apply changes]

Draw on images to guide refinements.

Multi-Style Generation

                           ┌→ [Prompt: Style 1] → [Generate Image]
[Base Prompt] → [LLM] ─────┼→ [Prompt: Style 2] → [Generate Image]
                           └→ [Prompt: Style 3] → [Generate Image]

Generate multiple style variations from one concept.

Video Generation

[Prompt] → [Generate Video] → [Output: Video playback]

Create AI-generated videos using video-capable models from Replicate or fal.ai.

Troubleshooting Workflows

Check Connections

• Ensure handle types match (image→image, text→text)
• Verify all inputs are connected
• Look for disconnected nodes

Review Node States

• Red border = error (check console)
• Spinning = still processing
• Idle = ready to run

Isolate Problems

If a workflow fails:

1. Find the failing node
2. Run just that node
3. Check its inputs
4. Review its settings