Image Generation & Editing

Generate and edit images with Nano Banana 2, Seedream, Flux, and FireRed.

Overview

YokeBot supports multiple image generation models and a dedicated image editing model. Agents choose the best model for the task, or you can specify one explicitly. All models are powered by fal.ai.

Generation Models

Model	Strengths	Credit Cost	Style Refs
Nano Banana 2	Fast, versatile, great default choice. Supports style references.	100	Yes (up to 6)
Seedream 3.0	Photorealistic, high detail, product imagery, portraits.	100	No
Flux	Artistic styles, creative compositions, illustrations.	100	No

Image Editing (FireRed)

The edit_image tool uses the FireRed Image Edit model to modify existing images based on text instructions. This is a separate tool from generate_image — use it when you need to change an existing image rather than create one from scratch.

Use cases:

Change backgrounds ("Replace the background with a mountain landscape")
Swap elements ("Change the red car to blue")
Add or remove objects ("Add a coffee cup on the table")
Style transfer ("Make this photo look like a watercolor painting")

Tool: edit_image
Parameters: prompt (required), image_url (required), aspect_ratio
Credit cost: 150

Style References

The generate_image tool supports style references via the image_urls parameter. Provide up to 6 existing images, and the model will use them to guide the visual style of the generated output while following your text prompt.

This is different from image editing — style references influence the overall aesthetic (color palette, composition style, visual mood) rather than modifying a specific image.

@design-agent Generate product photos for our new headphones.
Use these brand photos as style references for consistent lighting and background.
[attach 2-3 existing product photos]
Prompt: "Wireless headphones on a minimalist desk, soft natural lighting"

Parameters (generate_image)

Parameter	Type	Required	Default	Description
prompt	string	Yes	—	The text description of the image to generate.
aspect_ratio	string	No	1:1	Aspect ratio (e.g., "16:9", "4:3", "9:16", "1:1").
num_images	number	No	1	Number of images to generate (1–4).
image_urls	string[]	No	—	Up to 6 image URLs to use as style references.

Prompting Tips

Be descriptive: "A cozy coffee shop interior with warm lighting, wooden tables, and plants hanging from the ceiling" works better than "coffee shop".
Specify style: "digital illustration", "photorealistic", "watercolor painting", "isometric 3D render".
Include composition details: "close-up", "wide angle", "birds eye view", "centered".
Mention lighting: "golden hour", "studio lighting", "dramatic shadows", "soft diffused light".
For style references, describe what you want while letting the reference images handle the visual style.

Output

Generated images are saved to the workspace and posted inline in chat. Each image includes a thumbnail preview, the prompt used, and a download link. Files are automatically organized in the workspace file tree.

Batch Generation

Agents can generate multiple variants by setting num_images to 2–4. This is useful when exploring creative directions. The agent can then present all variants and let a human choose the best one via an approval workflow.