AI Uncrop Image

What Is AI Image Uncropping?

AI image uncropping — also called outpainting or generative fill — is the process of using artificial intelligence to extend an image beyond its original borders, generating new content that blends seamlessly with the existing photo. Unlike traditional image scaling (which stretches pixels and degrades quality) or canvas padding (which adds blank space), AI uncropping generates plausible new imagery that matches the colors, textures, lighting, and style of your photo.

This tool uses Stability AI's outpainting model to analyze the edges of your image and synthesize new content that fills the expanded canvas. The result is a larger image that looks like it was always that size — not stretched, blurred, or padded.

When to Use Each Expansion Direction

You can expand any combination of Top, Bottom, Left, and Right edges — each independently. Here is how to think about each direction:

• Top — Extends the sky, ceiling, or headroom above the subject. Useful when the original crop cuts off the top of a landscape, building, or portrait and you need more vertical space above the focal point.
• Bottom — Adds foreground detail — ground, floor, water, grass, or pavement below the existing image. Helpful for product shots that need more white space, or travel photos where you want more foreground context.
• Left / Right — Widens the image horizontally. This is the most common use case: converting a portrait or square crop to landscape for banners, headers, or widescreen formats. The AI extends whatever exists at the edges — trees, walls, street scenes, skies.

You can combine directions: for example, expanding left and right simultaneously transforms a square image into a 16:9 widescreen without distorting the center subject.

Using the Prompt Field

The prompt field is optional but often significantly improves results. When left empty, the AI infers what to generate purely from the existing image edges. When you add a prompt, the AI balances what it sees in the image with what you describe in text.

Good prompts are specific about content, not style:

• Good: "blue sky with light clouds", "dark hardwood floor", "blurred city street at night"
• Less useful: "beautiful", "photorealistic", "high quality"

The prompt works best when it describes something plausible given the original image content. If your photo is an indoor portrait with warm lighting and you ask to extend the right side, prompting "white painted wall with soft shadows" will yield a more coherent result than "mountain forest." Match the prompt to what logically belongs at the edges of your image.

Understanding the Creativity Slider

The creativity slider controls how strictly the AI adheres to the existing image edges versus generating more independently. Think of it as a dial between faithful extension andcreative generation:

• Low creativity (0.1–0.3) — The AI closely follows the colors, textures, and patterns at the edges of your image when generating new content. Use this for photos with strong, consistent backgrounds (solid walls, clear skies, uniform floors) where you want a very smooth, almost invisible extension.
• Medium creativity (0.4–0.6) — Balanced mode. The AI uses edge information as a strong guide but has enough freedom to generate coherent scene elements. This is the best default setting for most photographs.
• High creativity (0.7–0.9)— The AI treats the edges as inspiration rather than constraint, potentially generating more varied or unexpected content. Use this when the prompt describes something that needs to appear in the new space that isn't suggested by the edges (for example, extending into a dramatically different environment).

If the first result doesn't match your intent, try adjusting the creativity level and running the uncrop again — the model produces different outputs each run, so a few attempts often yields a significantly better result.

Changing Image Aspect Ratio with AI Uncrop

One of the most practical uses for AI uncropping is converting images between aspect ratios without losing the original content. Here are the most common conversions and how to achieve them:

• Square (1:1) to Landscape (16:9): If your image is 1000×1000px and you need 16:9, you need a 1778×1000 output. Expand left by ~389px and right by ~389px (or any combination totaling ~778px). Perfect for YouTube thumbnails, Twitter headers, and presentation slides.
• Portrait (4:5) to Square (1:1): Expand left and right evenly until the width matches the height. Instagram posts sometimes require square format when a portrait crop was used.
• Landscape to Ultra-Wide (21:9): Extend left and right by large amounts to create cinematic-width banner images for website headers and LinkedIn covers.
• Portrait to Landscape: Extend both sides equally to turn a vertical phone photo into a horizontal framing without losing the center subject.

Tips for Better Uncrop Results

• Start with clean edges. The AI extends what it sees at the edges. If your image has strong color gradients, clear textures, or consistent patterns at the borders, results will be better than if the edges are cluttered or have hard cuts.
• Expand one side at a time for large expansions. If you need to add more than 256px to a side, consider expanding 256px, downloading the result, and then running it again with another 256px. Smaller expansions per pass tend to be more coherent than very large single-pass expansions.
• Use lower creativity for uniform backgrounds. Blue skies, white walls, concrete floors, and similar surfaces extend most believably at low creativity values (0.2–0.3). The AI can reliably continue simple, consistent textures without needing much creative latitude.
• Use higher creativity with a guiding prompt.If your image has a complex background (a busy street scene, a forest, an interior with furniture) and you want to extend into new territory, set creativity higher and describe what should be there. The prompt gives the model direction that low creativity alone can't provide.
• Re-run for variation. The model generates a different result each time. If the first output has an awkward seam or unexpected element, click Uncrop again with the same settings for a fresh attempt. Most users find a good result within 2–3 tries.
• Avoid expanding people at the edges. If a person is positioned right at the edge of your image, extending in that direction may produce distorted body parts. Position the subject toward the center before using this tool, or use the prompt to describe neutral background content.

AI Uncrop vs. Other Resizing Methods

When you need a larger image, there are several approaches, each with different trade-offs:

• Scaling up (upscaling): Makes every pixel larger. Simple and fast, but produces blurry results above about 2× original size. Best when you need the same content at higher resolution, not a wider/taller frame.
• Canvas padding: Adds blank (usually white or black) space at the edges. The image itself is unchanged, but the extra space is empty. Useful for adding text margins, but not for filling in background content.
• Mirroring/reflection: Reflects the edge pixels outward. Creates a symmetric extension that often looks artificial. Works for abstract or decorative uses, not for realistic photos.
• AI outpainting (this tool): Generates new, plausible content that matches the existing image. Takes a few seconds and requires an API call, but produces results that look like the photo was originally taken with the wider frame. Best for realistic photographs and any case where the extended area needs to look natural.

Frequently Asked Questions