Quick Start

This tutorial is for a configured MCP client. For the web app, use Create your first image. Before either journey, connect a usable provider to your workspace; signing in alone does not supply provider access.

Your First Image

Open Claude Desktop, Claude Code, Codex, Cursor, or your MCP client and try this prompt:

Generate an image of a golden retriever playing in autumn leaves

Claude will use the generate_image tool and return a URL to your generated image.

TIP

The first request may take 3-5 seconds (Flash) up to 25 seconds (Pro / OpenAI high quality) depending on which provider you route to. Subsequent requests are typically faster, especially within the same session.

Customize Your Image

Specify Aspect Ratio

Generate a portrait image (3:4) of a majestic mountain landscape

Available aspect ratios:

• square (1:1) - Default
• portrait (3:4)
• landscape (4:3)
• 16:9 - Widescreen
• 9:16 - Vertical/mobile

Add Negative Prompts

Tell the model what to avoid:

Generate an image of a serene forest. Avoid: people, buildings, text, watermarks

Use System Instructions

Guide the overall style:

Generate an image of a coffee cup on a desk.
Style: Professional product photography with soft studio lighting

Edit Your Image

After generating an image, you can edit it conversationally:

Add some steam rising from the coffee

Make the background blurrier

Change the desk to dark wood

Each edit creates a new image while preserving the original.

Create a Character

Characters let you maintain consistency across multiple images:

Step 1: Create the Character

Create a character named "Captain Rex" - a robot explorer with a
bronze metallic body, glowing blue eyes, and antenna ears.
He wears a red scarf.

Step 2: Generate Scenes

Generate Captain Rex discovering an ancient temple in a jungle

Generate Captain Rex piloting a spacecraft through an asteroid field

Generate Captain Rex relaxing by a campfire under stars

The character maintains visual consistency across all scenes!

Create a Style Preset

Save your favorite style settings for reuse:

Create a style preset called "Vintage Film" with:
- Prompt prefix: "vintage 35mm film photograph, grain texture, muted colors"
- Negative prompt: "digital, sharp, oversaturated, modern"

Then use it:

Generate an image of a city street at night using the "Vintage Film" style

View Your Session History

Show me all images I've generated in this session

Or search across all your images:

Search for images containing "sunset" from the last week

Check Your Usage

Show me my usage statistics

What's my account summary?

Complete Example Workflow

Here's a complete creative workflow:

# 1. Create a character
Create a character named "Luna" - a young witch with long silver hair,
purple eyes, and a star-patterned cloak. She carries a wooden staff
with a glowing crystal.

# 2. Generate reference image
Generate Luna standing in her magical library

# 3. Add this as a reference image to the character
# (Claude will guide you through this)

# 4. Create scenes
Generate Luna brewing a potion in a moonlit forest

Generate Luna flying on a broomstick over a magical city

Generate Luna reading an ancient spell book by candlelight

# 5. Edit the last image
Add some floating sparkles around the candles

Make the atmosphere more mysterious

# 6. Check your work
Show me all images from this session

Tips for Better Results

Be Specific

❌ "Generate a cat" ✅ "Generate a fluffy orange tabby cat sleeping in a sunbeam on a window sill"

Include Context

❌ "Add clouds" ✅ "Add dramatic storm clouds with lightning in the background"

Use Art Terminology

Reference specific styles, techniques, and artists (styles, not copying):

• "Oil painting style with visible brushstrokes"
• "Minimalist vector illustration"
• "Cinematic lighting with lens flare"
• "Isometric 3D render"

Iterate Quickly

Start simple, then refine:

1. Generate base image
2. Edit to add details
3. Edit to adjust mood/lighting
4. Edit to fix any issues

Next Steps

Now that you've generated your first images:

• Learn about authentication and API keys
• Explore all features in depth
• Master MCP tools for precise control
• Set up the web console for visual management

Troubleshooting

"Tool not found" Error

Make sure your MCP client is properly configured and restarted after config changes.

Images Not Generating

Check the auth mode your client is using:

• Direct HTTP clients should be authenticated to https://mcp.gobananasai.com.
• STDIO proxy clients should have a local gobananas-cli auth login session or GO_BANANAS_API_KEY.
• All clients are still subject to quota and provider limits.

Slow Response Times

First requests may be slower. Gemini Pro and OpenAI gpt-image-2 are slower but higher quality than Flash; OpenAI high quality has the longest budget at 240s.

Character Not Consistent

Add reference images to your character for better consistency. The more reference images, the better.

Full troubleshooting guide →