Upscale Image
Upscale and enhance images using AI-powered super-resolution models
What does the Upscale Image node do?
The Upscale Image node increases the resolution of images using AI upscaling models. Choose from four different engines depending on your quality, speed, and creative needs.
Common uses:
- Enhance low-resolution product photos for e-commerce listings
- Upscale AI-generated images to print-ready resolution
- Prepare images for large-format printing or retina displays
- Restore old or compressed photos with improved detail
Quick setup
Add the Upscale Image node
Find it in AI Nodes → Upscale Image
Select an AI model
Choose the upscaling model (Topaz, AuraSR, Creative Upscaler, or ESRGAN)
Connect an image
Provide a single image as input (arrays are not supported)
Configure upscale settings
Set the upscale factor and model-specific parameters
Connect the output
Use the upscaled image in the rest of your workflow
Configuration
Required fields
model LLM selection required The AI model to use for upscaling. Each model maps to a different upscaling engine with distinct capabilities.
| Model | Best for | Max scale |
|---|---|---|
| Topaz | Professional photo upscaling | Varies by sub-model |
| AuraSR | Fast, lightweight upscaling | 4x |
| Creative Upscaler | AI art with creative enhancement | 4x |
| ESRGAN | Classic super-resolution, batch processing | 4x |
image image required The source image to upscale. Must be a single image — arrays are rejected.
upscale_factor number required Scale multiplier from 1x to 8x (maximum depends on the selected model).
- Start with 2x for most use cases
- Use 4x for significant enlargement
- Higher values produce larger files and take longer
Topaz
Professional-grade upscaling with 10 specialized sub-models.
Core sub-models: Standard V2, Low Resolution V2, CGI, High Fidelity V2, Text Refine
Generative sub-models: Recovery, Recovery V2, Redefine, Standard MAX, Wonder
face_enhancement boolean Enable face-specific enhancement with controls for strength, creativity, and sensitivity. Useful for portraits and group photos.
sharpen number Sharpen the upscaled image.
denoise number Reduce noise in the upscaled image.
fix_compression number Fix JPEG compression artifacts.
subject_detection boolean Enable automatic subject detection for targeted enhancement.
prompt string Text prompt for the Redefine generative sub-model. Guides the AI on how to enhance the image.
AuraSR
Fast and lightweight upscaling.
checkpoint string default: v1 Model checkpoint version: v1 or v2.
overlapping_tiles boolean Use overlapping tiles to reduce visible seams in the upscaled image.
Creative Upscaler
AI-enhanced upscaling with creative control for stylized results.
architecture string Base architecture: SD 1.5 or SDXL.
creativity number How much creative liberty the AI takes. Higher values produce more stylized results.
detail number Level of detail added during upscaling.
shape_preservation number How closely the output preserves the original shapes and structure.
guidance_scale number Guidance scale for the diffusion process. Higher values follow the prompt more closely.
negative_prompt string Describe what to avoid in the upscaled result (e.g., “blurry, artifacts, noise”).
ESRGAN
Classic super-resolution with multiple model variants.
variant string ESRGAN model variant.
| Variant | Description | Max scale |
|---|---|---|
RealESRGAN_x4plus | General-purpose, high quality | 4x |
RealESRGAN_x2plus | General-purpose, 2x only | 2x |
RealESRGAN_x4plus_anime_6B | Optimized for anime/illustration | 4x |
RealESRGAN_x4_v3 | Improved general-purpose | 4x |
RealESRGAN_x4_wdn_v3 | Better for noisy images | 4x |
RealESRGAN_x4_anime_v3 | Improved anime variant | 4x |
face_upscaling boolean Enable face-specific upscaling for better facial detail.
tile_size number Tile size for processing. Smaller tiles use less memory but may introduce seams.
Optional fields
output_format string default: PNG Output image format: PNG or JPEG. Some models have a fixed output format.
error_handling string default: None How to handle errors during upscaling.
None— Stop the workflow on errorSkip— Skip the node and continue
Output
The node outputs the upscaled image that can be connected to downstream nodes.
{
"url": "https://upscaled-image-url...",
"format": "png"
}Examples
Upscale product photos with Topaz
Use Topaz with the High Fidelity V2 sub-model at 2x to upscale product images while preserving fine detail. Enable sharpen and fix compression for images sourced from the web.
Enhance AI art with Creative Upscaler
After generating an image with Text to Image, connect it to Upscale Image using the Creative Upscaler with SDXL architecture. Set creativity to a moderate level and use a negative prompt like “blurry, artifacts, noise” to keep the output clean.
Batch upscale in a Loop with ESRGAN
Place the Upscale Image node inside a Loop to process multiple images. Use ESRGAN with the RealESRGAN_x4plus variant for a good balance of speed and quality across large batches.
Best practices
Choose the right model for your use case
- Topaz for professional photography and high-fidelity results
- ESRGAN for fast batch processing and general-purpose upscaling
- Creative Upscaler for AI-generated art that benefits from creative enhancement
- AuraSR for quick, lightweight upscaling when speed matters most
Start with 2x
Begin with a 2x upscale factor. Doubling resolution is usually sufficient and produces the most reliable results. Only go higher if you specifically need very large output.
Use face enhancement for portraits
When upscaling photos with people, enable face enhancement (Topaz) or face upscaling (ESRGAN) to preserve facial detail that generic upscaling can distort.
Common issues
Model not available
This feature is gated. Ensure your plan includes access to the selected upscaling model, and that a valid LLM selection is configured.
Image too large
Very large input images may exceed processing limits. Consider resizing the source image before upscaling, or use a lower upscale factor.
Blurry result
Try a different sub-model or variant. For Topaz, switch from a Core to a Generative sub-model. For ESRGAN, try the _v3 variants. Increasing the sharpen setting can also help.
Face looks wrong
Enable face-specific enhancement. In Topaz, adjust the face enhancement strength and sensitivity. In ESRGAN, enable face upscaling. Generative models may hallucinate facial details at high scale factors.