Geeky Ghost All In One Evolved — v4.1

Author notes from Geeky Ghost All In One Evolved v4.1.json · selected release

Fixed my cascade setup a bit. Wasn't working for some reason, so I made some adjustments based on another workflow. I add load clip and used the abc clip model I have, it was a cascade clip model just named model from somewhere. But I renamed it and have no idea where it originated lol. You can use clip from the stage c though if you don't have the original model for cascade clip. The PAG node seems to help with coherence a bit so I kept it. Stage C and B models from here. https://huggingface.co/stabilityai/stable-cascade/tree/main/comfyui_checkpoints Stage A model for VAE, but you can use B for the vae, there will be an error but it will still work. original workflow I borrowed bits from to merge with my own cascade group. https://civitai.com/models/409569/experimental-lcm-workflow-the-ravens-for-wurstchen-v3-aka-stable-cascade

Author notes from Geeky Ghost All In One Evolved v4.1.json · selected release

Image Generation Group is setup to do text to image with the empty latent Img2img with the latent switch and an uploaded image sketch to image if you'd like to attach the edit in another tab node. Lora stack and Controlnet Stack to allow you to select what you need. edit pose node available if needed. AIO adapters for image processing.

Author notes from Geeky Ghost All In One Evolved v4.1.json · selected release

Faceswap and Upscale for Video and Images Connect what you need, video ot image, simple upscale, faceswap, face correction Reactor isn't always easy to get up and running, feel free to swap with ip adapter faceID and run it through ksampler.

Author notes from Geeky Ghost All In One Evolved v4.1.json · selected release

Cascade feeds the image into SVD for the initial int making it text to image.

Author notes from Geeky Ghost All In One Evolved v4.1.json · selected release

AnimateDiff Vid2Vid group. This is designed to do vid2vid with images and video inputs for controlnet and loras to alter along with animateDiff I combine most things into this group so that you can alter videos anyway you want. It's fairly simple. Video is for controlnet. QRcode moster holds the shape, openpose can guide the pose including lipsync if you upload a video on yourself speaking. IP adapter takes the images and transfers the style and etc into the shape from controlnet. Added the multival dynamic if you'd like to play with it.

Author notes from Geeky Ghost All In One Evolved v4.1.json · selected release

The Mist fun node IMHO lol. Here you load 2 images, and blend them with the ip adapters, use controlnets and loras for even more shenanigans. Blend characters, scenes, styles, faces, etc.

Author notes from Geeky Ghost All In One Evolved v4.1.json · selected release

Layering Group! Main Subject 1st Background 2nd Secondary Subject 3rd I have it set up to remove the background from the subjects automatically and layer them on the background image. You can adjust the scale and position with the layering nodes. If you use the fast group bypass and bypass every group but this one, you can quickly adjust settings and perfect your layered int. Super hand and Helpful group. Can be used for wardrobe changes, action scenes, multi subjects, and etc. can even be used to layer special effects before running through img2img.

Author notes from Geeky Ghost All In One Evolved v4.1.json · selected release

Uses remove background to create a mask for the subject, then uses IP adapter to guide the background outpainting. Was able to extend out an additional 512 on each side pretty easily on my rtx 4090. Unorthodox solutions for the win lol.

Author notes from Geeky Ghost All In One Evolved v4.1.json · selected release

Uses IP adapter, inpainting specific nodes, Lora for LCM weights to speed up the generation, inpainting model Load Image, right click on image, select open in mask editor, and mask over what you want to replace. Prompt matters a bit as it helps it along.