ComfyUI Outpainting Guide: Extend Images Beyond Their Borders

What is Outpainting?

Outpainting extends an image beyond its original borders. The AI generates new content in the expanded area that seamlessly blends with the existing image. This is useful for:

• Expanding the field of view — turn a tight crop into a wider scene
• Fixing composition — add headroom, extend a landscape, or fill missing edges
• Changing aspect ratio — convert a square image to widescreen or portrait

Outpainting works on the same principle as inpainting — the extended area is treated as a masked region that the AI fills in.

Prerequisites

You need a checkpoint model. For best results, use a dedicated inpainting model:

Place both in ComfyUI/models/checkpoints/.

Building the Outpainting Workflow

The workflow has two stages:

Stage 1: Prepare the Expanded Canvas

1. Load Image — load the image you want to extend
2. Image Pad for Outpaint — adds blank padding around the image

The Image Pad for Outpaint node has four parameters controlling how many pixels to add in each direction:

Set the directions you want to extend. For example, to make a landscape wider, increase left and right. The node also generates a mask for the padded area.

Stage 2: Generate the Extended Content

3. Load Checkpoint — load an inpainting model
4. VAE Encode (for Inpainting) — encodes the padded image and mask into latent space
5. CLIP Text Encode (positive) — describe the scene (should match the original image's content)
6. CLIP Text Encode (negative) — elements to avoid
7. KSampler — generates content in the masked area
8. VAE Decode → Save Image

Key Principle

Use the same prompt style as the original image. If the original is a landscape photo, prompt for landscape elements. Consistency between the prompt and the existing image is what makes the extension look natural.

Tips for Natural-Looking Results

Keep prompts consistent — The extended area should match the original image's style, subject, and mood. Don't prompt for completely different content.

Use an inpainting model — Dedicated inpainting models handle edge blending much better than standard models.

Start small — Extend by 128–256 pixels at a time rather than trying to double the image in one pass. Multiple small extensions produce more coherent results.

Adjust KSampler parameters:

Common Issues and Fixes

Visible seam between original and extended area

• Use a dedicated inpainting model
• Lower cfg to 5–7
• Extend in smaller increments
• Adjust the feathering parameter on the pad node if available

Extended area doesn't match the original style

• Rewrite your prompt to describe the entire scene, not just the new area
• Use similar quality keywords as the original image
• Try a different sampler (euler_ancestral often blends well)

Extended area is blurry compared to the original

• Increase steps to 30+
• Increase the resolution of the padded canvas
• Consider running a second pass with inpainting to refine details

Colors don't match between original and extended areas

• The inpainting model typically handles color matching well — if colors diverge, try a different model
• Add color/lighting keywords to your prompt that match the original

Related Guides

• Inpainting Guide — Selectively edit parts of an image
• Image to Image — Transform existing images
• Upscale Guide — Enlarge images with AI detail enhancement