What denoise value should I use for ComfyUI upscaling?

For latent upscaling (hi-res fix), keep denoise below 0.5 to preserve your original composition. Lower values (0.2–0.35) add sharpness without changing the image much. Higher values (0.5+) let the model reimagine details, which can sometimes improve quality but may alter the image significantly.

Why does my upscaled image look blurry or have artifacts?

Blurry results usually mean the upscale factor is too high or the denoise is too low. Artifacts like blockiness or ringing are often caused by over-sharpening or using a model incompatible with your image type. Try reducing the upscale factor, switching to a different upscale model (e.g. Realesrgan vs latent), or adjusting the denoise and sharpen settings.

ComfyUI Upscale: Beginner Guide to Image Upscaling (2026)

Why Upscale Images?

Quick answer: Most image-generation models produce 720p–1080p output. Industries like printing, large-format displays, and professional AI art require 4K or 8K resolution. ComfyUI upscaling lets you generate a detailed base image and then intelligently enlarge it while preserving or enhancing detail.

The typical ComfyUI workflow is to generate a well-composed image at 720p–1080p, then use upscaling tools to reach 4K or 8K — while keeping fine detail intact. Here are the main reasons to upscale:

Photography, Prints, and Social Media

Upscaling helps produce print-ready images with less visible pixelation and richer detail at large output sizes. For social media, upscaling turns low-res exports into crisp, high-quality thumbnails and post images.

Photography upscaling — a low-resolution image enlarged to print quality using ComfyUI

Social media upscaling — improving image clarity for thumbnails and posts

AI Art and Image Restoration

AI art: Upscaling can enhance style and add fine detail while keeping the original composition intact.
Restoration: Improve old or damaged images by intelligently reconstructing faces, textures, and small details that were lost due to compression or age.

Image restoration using ComfyUI upscaling — reconstructing detail in a degraded photograph

What Is ComfyUI Image Upscaling?

ComfyUI upscaling means converting a low-resolution image into a higher-resolution version using the ComfyUI node interface. You can use simple pixel interpolation methods or AI models that intelligently reconstruct missing detail.

Core Upscaling Methods

Pixel Upscaling

Directly increases visible resolution by duplicating or interpolating pixels. Good for small resolution changes.

Best for: Small scale-ups where speed matters more than quality

Traditional Interpolation

Bicubic, Bilinear, or Lanczos algorithms. Fast but can look soft or blurry at large scales. Available via the Upscale Image By node.

Best for: Quick previews and non-critical output

AI Model Upscaling

Uses models like Realesrgan, SwinIR, or 4x-UltraSharp to intelligently rebuild detail. Uses Load Upscale Model and Upscale Image (using Model) nodes.

Best for: High-quality final output for art and photography

Latent Upscaling (Hi-Res Fix)

Upscales inside the latent space with a second sampler pass using the Upscale Latent node. Can add new creative details — keep denoise below 0.5 to preserve composition.

Best for: Adding fine AI-generated detail to existing images

Quick Note on Terms

Upscaling = increasing resolution while reconstructing or adding detail.
Enhancement = improving perceived quality through denoise, sharpening, face restoration, or color correction.

Popular Upscaling Methods and Models

There are several proven upscaling workflows for ComfyUI. Each has different strengths — choose based on the quality level you need and the type of image you are working with.

Using an SDXL Model for Upscaling and Face Refining

This method uses a full SDXL checkpoint model for latent upscaling combined with face refinement. It produces high-quality results with sharp detail and natural-looking faces — especially useful for portrait upscaling.

Download SDXL Upscale Workflow

Download and open the workflow.

Save the JSON file above and drag it into ComfyUI to load the workflow.

Fix any red nodes if they appear.

If you see red nodes, follow the missing nodes fix guide: How to Install Missing Nodes in ComfyUI

Add the image you want to upscale.

Use the Load Image node to select your source image. Click the node and upload your file.

ComfyUI SDXL upscale workflow showing the Load Image node where you add the file to upscale — Add the image you want to upscale using the Load Image node in the SDXL upscale workflow.

seedvr2

seedvr2 is useful for specific artistic detail profiles and produces distinctive texture and sharpness characteristics. Try seedvr2 when you want a different texture style from the standard SDXL latent method.

Download seedvr2 Workflow

seedvr2 upscaling workflow in ComfyUI showing the node layout for artistic texture-preserving upscaling — The seedvr2 workflow in ComfyUI — good for specific artistic detail styles.

seedvr2 upscaling result — enhanced texture and detail in the upscaled output

Ultimate SD Upscale / 4x-UltraSharp

Designed for Stable Diffusion pipelines, Ultimate SD Upscale is a classic tile-based upscaling method. It is older now and has largely been surpassed by newer latent and AI model methods — but it can still produce good results on SD 1.5 pipelines and is worth knowing.

Download Ultimate SD Upscale Workflow

Ultimate SD Upscale workflow in ComfyUI showing tile-based upscaling node configuration with 4x-UltraSharp model — The Ultimate SD Upscale workflow — tile-based upscaling for Stable Diffusion pipelines.

Settings & Practical Tips

These four settings have the biggest impact on your upscaling results. Understanding each one will help you get clean, sharp output without artifacts.

Denoise

Lower denoise preserves the original composition; higher denoise lets the model reimagine details. For latent upscaling (hi-res fix), keep denoise generally below 0.5. A value of 0.2–0.35 is a good starting point for most upscaling tasks.

Face Restoration

Face restore nodes improve portrait quality significantly. Use them sparingly — overuse can make faces look unnatural or plastic. Start at a low strength and increase gradually, testing after each change.

Sharpening

Apply sharpening after upscaling if the result looks soft. Small amounts are usually enough — too much sharpening introduces halo artifacts around edges, which is hard to reverse.

Artifacts

Common artifacts include blockiness, ringing, and over-smoothing. To reduce them: try a different upscale model (e.g. Realesrgan vs latent), reduce the upscale factor, or adjust denoise and sharpen settings.

💡 Tip: Always test your upscaling settings on a small crop or low-resolution version of your image first. Once the settings look right, run the full-resolution upscale. This saves significant processing time.

Batch Processing Multiple Images

If you need to upscale many images with the same settings, use a batch upscaling workflow. This saves you from manually running the workflow for each file.

Download Batch Upscale Workflow

Place all images in one folder.

Organise your source images into a single directory before running the batch.

Use a folder-reading node to loop the workflow.

The batch workflow reads each file in the folder and processes them sequentially.

Test with one image first.

Confirm your settings are correct on a single image before running the full batch — this avoids wasting time on incorrect settings.

ComfyUI batch upscaling workflow showing folder input node for processing multiple images at once — The batch upscale workflow — process an entire folder of images with one run.

Combining Text-to-Image Workflows with Upscaling

The most efficient ComfyUI workflow generates and upscales in one pass: your text-to-image nodes feed directly into the upscaling nodes. This eliminates the need to save, reload, and re-import images manually.

Download Combined T2I + Upscale Workflow

Link your text-to-image nodes (KSampler → VAE Decode → Image) directly to the upscaling input. The image is generated and then immediately passed through the upscaler in a single queue run.

ComfyUI text-to-image combined with upscaling workflow — nodes connected from KSampler through to the upscale output — A combined text-to-image + upscaling workflow — generate and upscale in a single run.

💡 Tip: Use the ComfyUI Text-to-Image guide to set up your generation nodes correctly before connecting the upscaling stage.

Beginner Mistakes to Avoid

These are the most common errors people make when first using ComfyUI upscaling — and how to avoid them:

Upscaling noisy or heavily compressed images without cleaning first.

Fix: Apply a denoise or cleanup pass before upscaling. Feeding garbage in produces garbage out — upscalers amplify existing noise and compression artifacts.

Using very high denoise values with latent upscaling.

Fix: High denoise (above 0.6) causes the model to reimagine the image entirely, losing your original composition. Keep denoise below 0.5 for upscaling tasks.

Upscaling full-resolution images without testing on a crop first.

Fix: Always test settings on a small cropped section of the image. This gives you fast feedback on whether the settings are correct before spending time on a full-resolution run.

⚠️ Warning: If you see red nodes when loading any of the workflows above, you are missing a custom node. Follow the missing nodes installation guide to fix this before trying to run the workflow.

Frequently Asked Questions

Image upscaling in ComfyUI means converting a low-resolution image into a higher-resolution version using AI models or interpolation methods. You can use nodes like Upscale Image (using Model), Upscale Latent, or Ultimate SD Upscale to increase resolution while preserving or adding detail.

The best method depends on your use case. For highest quality AI upscaling, seedvr2 and SDXL-based latent upscaling produce excellent results. For fast, reliable upscaling with good detail, 4x-UltraSharp or Realesrgan AI models work well. Ultimate SD Upscale is a classic option that still works but has largely been surpassed by newer methods.

For latent upscaling (hi-res fix), keep denoise below 0.5 to preserve your original composition. A value of 0.2–0.35 is a solid starting point. Higher values (0.5+) let the model reimagine details, which can improve quality but may significantly alter the image.

Use a batch upscaling workflow. Place all your images in one folder, use a Load Image node that reads from that folder, and loop the workflow to process each file. Download the batch upscale workflow provided in this guide to get started immediately.

Blurry results usually mean the upscale factor is too high or denoise is too low. Artifacts like blockiness or ringing are often caused by over-sharpening or using a model that is incompatible with your image type. Try reducing the upscale factor, switching to a different model (e.g. Realesrgan vs latent), or adjusting the denoise and sharpen values.