Basic Outpainting Workflow

Outpainting shares similarities with inpainting, primarily in that it benefits from utilizing an inpainting model trained on partial image data sets for the task.

This is because the outpainting process essentially treats the image as a partial image by adding a mask to it.

I demonstrate this process in a video if you want to follow along:

1. Image Padding

The initial step in ComfyUI involves padding your original image using the Pad Image for Outpainting node, accessible via Add Node > Image > Pad Image for Outpainting. This functionality allows for expansion in any direction and includes an option for feathering the edges of the source image, which can help blend the new sections smoothly with the original.

The maximum allowable feathering is capped based on the image's dimensions, ensuring the effects remain proportionate and visually coherent.

2. Image Encoding

Following the setup of the padding parameters, the image undergoes encoding through the VAE Encode (for Inpainting) node (Add Node > Latent > Inpaint > VAE Encode (for Inpainting)). This stage converts the image and its mask into a latent space representation, which is crucial for outpainting.

You can typically leave the grow_mask_byvariable at its default value.

3. Utilizing Prompts & KSampler

The next phase involves guiding the outpainting process through positive and negative prompts, steering what the model generates based on the latent image context.

Best results are to describe the current image and what you want it to look like after outpainting.

Be sure to set the KSampler values to the correct settings for the model you are using.

4. Evaluating the Results

Once set, click generate and review the results! You can adjust the prompts and KSampler settings as needed to refine the output.

👉 Consider checking out the Prompting Pixels for other helpful guides, workflows, tools, and more.