ControlNet Pose

ControlNet Pose is a model that modifies images by using human pose information as an input condition when generating imagery. The model combines a pose map with a text prompt to guide image synthesis, using a diffusion-based approach. It presents what the model does, who might use it, and how it integrates with Replicate’s API and infrastructure.

It explains how ControlNet extends pretrained diffusion models to support additional input conditions beyond prompts, and notes the potential for training and running on personal hardware or clusters.

The model includes a description of the general ControlNet approach, including support for inputs beyond pose maps, such as edge maps and keypoints. It notes that the technique can scale with computation resources, and that ControlNets can augment large diffusion models to enable conditional inputs. It also references the availability of multiple options for using ControlNets to modify Stable Diffusion outputs.

It provides context for where the model fits in the broader family of ControlNets and how it relates to the Stable Diffusion ecosystem, including the option to run it locally with Docker and considerations for hardware like Nvidia GPUs.

Use cases

• Use pose-based control to modify Stable Diffusion outputs