Casual/everyday activity poses for controlnet.

I am open to suggestions for what other categories to add to this so-far boring list.

Please read the version description for notes that are specific to the version.

In the .zip files: example generation plus the .png to use in openpose, canny and depth models. Added .json for openpose too so you can make modifications to the pose file. They're not always very functional though.

To do list

• Sleeping outside of bed (slumped in a chair, head down on a table, curled up on the floor) -- suggested by 0ryan -- DONE

• Interacting with pets (petting a cat, walking a dog, feeding fish, lizard on shoulder) -- suggested by 0ryan

• Playing games (holding a controller, moving a chess piece, dealing cards, etc) -- suggested by 0ryan

• Yoga

• Reading

• Office/computer work

• ...

Additional notes

• Use attributes like "from_behind", "from_below", "from_above", or "from_side" in your prompt to create the angles you are looking for. Describe accurately what you want to see (such as "stomach", "thighs", "back" etc.) to help "turn" a character.

Version notes

• As always, aspect ratio is crucial to the output.

• Phone: Use canny for most flexibility. Canny fills the blank spots very well and you can easily edit the file to make changes to the template. Openpose (full) also works pretty well in this one as it helps getting the fingers right. The .json is practically useless for this version. If you want to generate something that is close to the example picture (an athletic woman with long hair), use depth or run the example picture through the normal map preprocessor and model. Useful keywords: smartphone, holding smartphone, selfie, phone, mirror selfie, etc.

• Dozing: Would suggest a combination of OpenPose (weight 1.0) and depth (weight 0.3-0.8) in multi ControlNet here. OpenPose alone is not enough in most cases. Depth alone might suffice in most cases though.

• Presenting: I included "(presenting a blank_page:1.5)" in my prompts and it usually got the job done in keeping the object a blank white space (but responding to the lighting of the scene). Pose 1 is a little tricky to pull off. Try not to describe the upper body of the character holding the poster/painting/document too much as parts of it would then show up ON the object. Some odd looking results. You can combine depth and canny (to improve fingers) for pose 1. Depth is doing the heavy lifting here though. With just canny you will not get the right results. The contents of the presentation can easily be generated/modified via inpainting. Paint a mask over the poster/painting/photo, "only masked" instead of "whole picture", 512x512 and a denoising strength of 0.6 to 0.75 usually gets the job done.

All pictures were generated with the Comimicry model (https://civitai.com/models/105172). DPM++ 2M Karras sampler.

Lora used: 3D Rendering Style (https://civitai.com/models/73756/3d-rendering-style) v1.4.1

Prompt and other generation data: see attached pictures.