✳flâneur — a map of the web's best reading

ComfyUI_Workflows/image_conditioning/README.md at main · cubiq/ComfyUI_Workflows · GitHub

github.com · 1,086 words · saved by 1 readers

A picture is worth a thousand words, sometimes starting from a reference image is more efficient than trying your luck with simple text-to-image. In this section we'll explore various image-to-image techniques. If you want to follow the following examples be sure to download the content of the input directory of this repository and place it inside ComfyUI/input/. The easiest of the image to image workflows is by "drawing over" an existing image using a lower than 1 denoise value in the sampler. The lower the denoise the closer the composition will be to the original image. We can of course augment the generation with proper prompting. In this workflow we use a base image of a portrait of a woman to create a similar image of a man. 👉 Note: We are using SDXL for this example. The latent size is 1024x1024 but the conditioning image is only 512x512. It is a good idea to always work with images of the same size. That's why in this example we are scaling the original image to match the late

Image-to-Image Conditioning Simple Img2Img unCLIP model Style Model IPAdapter image + text SDXL Revision Experiments unCLIP multiple images unCLIP with SDXL refiner augmentation IPAdapter image variations IPAdapter + Canny control net Timestepping a Style model Image-to-Image Conditioning A picture is worth a thousand words, sometimes starting from a reference image is more efficient than trying your luck with simple text-to-image . In this section we'll explore various image-to-image techniques. If you want to follow the following examples be sure to download the content of the input director

Explore this link on the map →

related reading