
This workflow for ComfyUI takes a subject image, does a simple background remove, and overlays it onto a background. Both subject and background can be tuned with loras, both have controlnets that can be bypassed in groups. Right now it's set up for PonyXL but you can plug any model into it, just be sure to change ALL the controlnets/loras to appropriate models.
The image composite area has X and Y controls, plus scaling of the foreground subject. I recommend doing some initial tests to get the positioning right. There's a global seed for all the ksamplers set that you change to randomize or fixed.
After compositing the images, a composite depth-map is created, so that the Hiresfix pass to merge the images preserves the depth. There's some image controls to adjust the depth mask (using contrast and brightness primarily to tweak, you want the foreground subject to be brighter overall than the background depthmask).
Next step is Ultimate Upscale - if you're using a 1.5 model, be sure to change the tile size down to something more appropriate. Adjust the scaling as you like, but I recommend between 1.5x to 2.5x at this step. Adjust the denoise value higher the greater you scale it (from 0.2 to 0.3, depending on your model and loras).
Last step is FaceDetailer - there's an optional Lora you can apply to facedetailer, if you're trying to make a particular character, for example.
There are optional groups that are bypassed by default that you can reactivate and add back in:
- Subject Background remove - it also adds a color background to the subject, useful if you have a busy source image.
- IC lIght - by default it simply uses the composited image to relight, which does a burn effect, you will have to manually add in your own lighting map if you want to change it.
- Handfix - disabled by default because it takes a lot of futzing to make it work right
描述:
-
initial release Pony version
initial release Pony version
训练词语:
名称: subjectBackground_v10.zip
大小 (KB): 20
类型: Archive
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success