We are making an Agent that can automate the whole personalized visual content creation process. It will need to perform multiple types of tasks, including designing tasks and training a model for it's own use. This agent will handle the data, the models and ensure the quality of the outputs.

Here are some randomly picked weights we trained on image-conditionned text to image generation for spatially aligned and non-aligned tasks.

We are training multiple tasks in various conditions and settings. We will share more details soon on the weights and how to run them, but these weights were finetuned with versions of OminiControl Pipelines. Most, if not all should work with OminiControl so you can add them to your project and load them as adapters. the adapter name is the string in the folder name after "-".

Here are the Controls, Tasks and Categories we are taining for and we plan to make them all open source, including the video models.

Controls:

Preprocessing - bg remove, matting, reshaping, segmentation… Control Models - Shape (canny, Hed, scribble, depth..), Pose, Mask, Camera view Post-Processing models - Enhancement, Color fixing, blending Editing Models - in painting (including removal, masked blending, replacing etc…, adding), Outpainting, Motion/Transformation , Relighting

Tasks:

Background generation Controlled Background Generation Subject consistent in context Generation Object placement Video Generation In context Image/Video Generation multi-object/subject Merging/Blending

Categories:

Product photography Fashion Accessory / Shoes, Hats etc…. fitting Virtual Try-on People