Image Generation Nodes

Generate and upscale images using AI models.

Text to Image

Generate images from text descriptions using AI models

The primary node for AI image generation. Transforms text prompts into images, with optional edit mode for modifying existing images.

Property	Description
Inputs	ctx (text), images (image, optional -- required for edit models)
Outputs	images (image)

Configuration:

Setting	Description
Model	AI model for generation (see table below)
Aspect Ratio	Output aspect ratio (e.g., 16:9, 1:1, 9:16)
Resolution	Output resolution (e.g., 1024*1024, 2048*2048, 2k, 4k)
Seed	Random seed for reproducibility (-1 for random)
Negative Prompt	What to avoid in generation

Available Generation Models

Model	Key Features
google/nano-banana-pro/text-to-image	Configurable aspect ratio and resolution (default 2k), JPEG output
google/nano-banana-pro/text-to-image-ultra	Ultra-quality variant, PNG output, up to 4k resolution
google/nano-banana-2/text-to-image	Second-generation Nano Banana, JPEG output
bytedance/seedream-v4.5	Budget-friendly, configurable size (default 2048x2048), supports seed
bytedance/seedream-v5.0-lite	Lightweight SeedDream v5, configurable size
wavespeed-ai/flux-kontext-dev-lora-ultra-fast	Context-aware generation, supports LoRA adapters, configurable guidance scale and inference steps

Available Edit Models

Edit models require an image connected to the images input. They modify the input image based on the text prompt.

Model	Key Features
bytedance/seedream-v4.5/edit	Image editing with SeedDream
bytedance/seedream-v4/edit-sequential	Sequential multi-image editing, configurable max images
bytedance/seedream-v5.0-lite/edit	Lightweight SeedDream v5 editing
google/nano-banana-pro/edit	Nano Banana Pro editing, PNG output, 1k resolution
google/nano-banana-pro/edit-ultra	Ultra-quality editing, PNG output, up to 4k
google/nano-banana-2/edit	Second-generation Nano Banana editing

How It Works

1. The handler validates the prompt text and determines if the selected model is an edit model (model path contains /edit).
2. Edit models require an image input; standard models optionally accept reference images.
3. The request is sent to the generate-image edge function, which routes to the WaveSpeed API.
4. For async jobs, the node polls for completion with progress updates displayed in the editor.
5. On success, the generated image artifact is stored in the flow context.

If a job fails or times out, credits are automatically refunded via CreditRefundError.

---

Image Upscale

Increase image resolution using AI

Upscales images to higher resolutions using the WaveSpeed Ultimate Image Upscaler.

Property	Description
Inputs	image (image)
Outputs	output (image)

Configuration:

Setting	Description
Model	Upscale model (default: wavespeed-ai/ultimate-image-upscaler)
Target Resolution	Output resolution target (default: 4k)
Output Format	Image format (default: jpeg)

How it works:

• Sends the input image to the generate-image edge function with the upscaler model path
• Polls for async job completion with progress updates
• Returns the upscaled image as an artifact

Use cases:

• Prepare images for print or large displays
• Enhance low-resolution source images
• Create 4K content from smaller images

---

Tips for Best Results

Prompt Quality

Write detailed prompts for better generation:

"Professional headshot of a business woman,
 studio lighting, neutral gray background,
 sharp focus, high-end camera quality"

Avoid vague prompts like "Woman photo".

Aspect Ratios

Choose based on use case:

Ratio	Use Case
1:1	Social media posts, profiles
16:9	Landscape, presentations
9:16	Stories, vertical video
4:3	Traditional photography

Model Selection

1. Nano Banana Pro -- Fast, high-quality general-purpose generation
2. SeedDream v4.5 -- Budget-friendly generation (used in Budget Mode)
3. Flux Kontext -- Context-aware generation with LoRA support
4. Edit models -- When modifying existing images with text instructions