アウトペインティング：画像の境界を拡張する

Outpainting: Expanding image boundaries | Runware Docs Introduction Outpainting allows you to expand images beyond their original boundaries , extending scenes in any direction while maintaining visual continuity. Unlike inpainting which modifies existing areas, outpainting generates new content that seamlessly connects to your original image edges. At its core, outpainting is a specialized form of inpainting that operates on a mask created outside the original image dimensions. While you can manually create and provide this mask, our API offers a streamlined approach with the outpaint parameter that automatically generates the appropriate mask based on your specified directions and dimensions. This guide covers everything you need to know about outpainting with the Runware API, from basic concepts to advanced techniques for creating natural-looking expansions. Basic request example Here's a simple outpainting request to get you started: Request [ { "taskType" : "imageInference" , "taskUUID" : "c985ec13-6aef-4337-8df2-f7e17dd71589" , "positivePrompt" : "__BLANK__" , "seedImage" : "c985ec13-6aef-4337-8df2-f7e17dd71589" , "outpaint" : { "top" : 256 , "right" : 256 , "bottom" : 256 , "left" : 256 } , "model" : "runware:102@1" , "width" : 1280 , "height" : 1280 , "steps" : 40 } ] Response { "data" : [ { "taskType" : "imageInference" , "imageUUID" : "5ba056f3-3d09-4209-93c4-334df0533878" , "taskUUID" : "b2728981-a3bd-45ae-b263-a6accde10332" , "seed" : 6110856530272051867 , "imageURL" : "https://im.runware.ai/image/ws/2/ii/5ba056f3-3d09-4209-93c4-334df0533878.jpg" } , ] } This request extends our original landscape image by 256 pixels in every direction. The model generates appropriate content in these new areas based solely on the original image context . While a detailed prompt can be provided for guidance, here we've used __BLANK__ to allow the model to extend the scene based entirely on the visual information from the original image edges. How outpainting works Outpainting extends your image by creating new content beyond the original boundaries that integrates seamlessly with existing elements. At its technical core, outpainting is a specialized form of inpainting that operates on a mask positioned outside the original image dimensions . When you submit an outpainting request, the API first creates a larger canvas based on your specified dimensions. This is why it's critically important to set the width and height parameters to reflect the final combined dimensions (original image size plus extensions). Your original image is then positioned on this canvas, leaving empty spaces in the areas to be generated according to your outpaint directions. The technical process begins with what's called edge extension or color bleeding , where the system extends the edge pixels of your original image outward to create initial guidance in the new areas. This step is necessary because unlike regular inpainting where the model has actual image content under the mask as a starting reference, in outpainting these new areas begin as undefined canvas space with no existing pixel data . Prompt : A butterfly wing under magnification, revealing scales that shimmer like tiny stained glass windows in rainbow hues. Original (768x768) Canvas (384+128+128+384) Mask layout (1280x1280) From this extended starting point, the generative model analyzes patterns, colors, textures, and structural elements at the boundaries of your original image. It uses this contextual information along with your prompt guidance to determine what content should be generated in the extended regions . The strength parameter controls how much of the initial edge extension is preserved versus replaced by prompt-guided content. Final The boundary transition between original and generated content is controlled by the optional blur parameter. This parameter creates a gradient in the mask at the boundary between the original image and new areas. With higher blur values, the transition becomes more gradual, allowing the original edge pixels and their extensions to blend more smoothly with the newly generated content. This helps avoid abrupt transitions, particularly useful when extending complex patterns or textures. What makes outpainting particularly challenging is that the model must generate coherent extensions with only partial context available from the image edges. Unlike standard inpainting where the model can reference surrounding content from all directions, outpainting requires the model to extrapolate beyond what's visible, effectively "imagining" how the scene continues based on limited edge information . Outpaint level 0 Key parameters Outpaint: Defining the expansion The outpaint parameter is an object that specifies which directions to expand your image and by how much. This is the core parameter that activates the outpainting functionality. These extension values define the exact pixel dimensions of the new canvas regions that will be created around your original image. The model's generation process will be confined to precisely these regions, while the original image remains untouched. The blur parameter deserves special attention as it creates a gradient mask transition at the boundaries. Higher blur values create a more gradual blend that can help disguise transitions in complex textures or patterns. Lower values maintain sharper edges, which can be preferable when extending architectural elements or when precise structural continuity is essential. Outpaint Blur 0 When applying different blur values, you can observe how this parameter affects content continuity at boundaries . Higher blur values create more seamless transitions between original and extended areas, maintaining stronger visual continuity. This is particularly noticeable in the river, where increased blur helps preserve its natural flow and characteristics across the boundary between original and generated content. You can expand in any combination of directions simultaneously . For example, you might expand only to the right, or in all four directions at once. The requirement for extensions to be multiples of 64 pixels is related to how image generation models process data in latent space. Using these dimensions ensures optimal performance and quality. Dimensions: Critical for outpainting When using outpainting, the width and height parameters take on special importance. They must explicitly account for the combined dimensions of your original image plus the extensions. For correct outpainting, calculate these values as: width = Original image width + left extension + right extension. height = Original image height + top extension + bottom extension. If you don't specify the correct dimensions that account for both the original image and extensions, your outpainting result may be unexpected. Always calculate and provide accurate width and height values. Strength: Understanding the critical threshold The strength parameter controls how much influence the original image edges have on the newly generated areas during outpainting. This parameter works quite differently for outpainting compared to standard image-to-image tasks. When outpainting begins, the system first creates what's essentially a color bleed extension from the edge pixels of your original image. These extended color values serve as initial guidance for the generation process. The strength parameter determines how much of this initial guidance is preserved versus replaced by prompt-guided content. At low strength values (0.0-0.6), the result is dominated by this color bleeding effect, creating extensions that are essentially just stretched colors from the edge pixels without meaningful detail or content. These low values typically produce unusable results for most outpainting purposes. Effective outpainting typically starts at strength values of 0.7 and above: Values around 0.7-0.8 maintain significant influence from edge patterns while allowing new content. Values around 0.9-0.95 provide an optimal balance for most outpainting scenarios, offering creative freedom while maintaining reasonable continuity. Values at or very near 1.0 may cause subtle variations in the original image portion, though these changes are typically minor. Strength 0 For most outpainting tasks, we recommend strength values between 0.85-0.95 for the best balance between creative generation and edge consistency. This range gives the model enough freedom to generate meaningful content while still respecting the visual cues from your original image boundaries. FLUX Fill and strength parameter FLUX Fill does not use the strength parameter as it employs a different technical approach to outpainting. When using FLUX Fill ( runware:102@1 ), the model automatically determines the optimal balance between edge guidance and new content generation, making the strength parameter unnecessary. Simply omit this parameter when using FLUX Fill for outpainting. Seed image: The starting point The seedImage provides the base image to be extended. Unlike standard inpainting, with outpainting this image will be positioned within a larger canvas according to your expansion directions. The original image is kept intact, with new content generated only in the expanded areas. As mentioned earlier, the system first extends the edge pixels of your seedImage outward to create initial color guidance in the new areas . This edge extension provides the starting foundation upon which the model builds the generated content. The seedImage quality and content significantly influence the outpainting result, as the model draws contextual cues from the image edges . Other critical parameters The positivePrompt parameter guides what content should appear in the expanded areas . As with inpainting, you can focus primarily on describing what should appear in these new regions. Since the model has access to the entire original image in outpainting, it can understand the existing scene contex