I haven't checked, but perhaps something similar can be done in SDv2. To use Inpaint, first you drag a photo onto the Inpaint icon or choose File > Open, or click the Open button in the toolbar, and then choose an image. And it worked very well! You can see the result and parameters of inpainting in the screenshots.ġ Go to Checkpoint Merger in AUTOMATIC1111 webuiĢ Set model A to "sd-1.5-inpainting" model ( )Ĥ Set model C to "v1.5-pruned" model ( )Ħ Choose "Add difference" Interpolation methodħ Make sure your model has the "-inpainting" part at the end of its name (Anything3-inpainting, DreamLike-inpainting, etc.) I decided to try using the "Add difference" option and add the difference between the 1.5-inpainting model and the 1.5-pruned model to the model I want to teach the inpainting.
If you just combine 1.5 with another model, you won't get good results either, your main model will lose half of its knowledge and the inpainting is twice as bad as the sd-1.5-inpainting model. Evaluations with different classifier-free guidance scales (1.5, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0) and 50 steps DDIM sampling steps show the relative improvements of.
Use it with diffusers Model Details Developed by: Robin Rombach, Patrick Esser.
#Inpaint 3.0 free download download
We already have sd-1.5-inpainting model that is very good at inpainting.īut what if I want to use another model for the inpainting, like Anything3 or DreamLike? Any other models don't handle inpainting as well as the sd-1.5-inpainting model, especially if you use the "latent noise" option for "Masked content". Use it with the stablediffusion repository: download the 512-inpainting-ema.ckpt here.
0 kommentar(er)
