near plover
Two similar but different HRO node graphs mostly parameterised.
They both use the same parameter setup on the left side of the graph and include an add_detail lora node which I feel can improve upscaling.
- Model Loader
- Steps
- CFG Scale
- Height
- Width
- Scale factor
- Rand Range (number of images to produce)
-
The first (HRO-Resize-L2L-RandRange-Param-Example.json) uses Image Resize, L2L and a tile control net. So it is a bit like an Image to image upscale with a control net done in one shot. You get the benefit of a strength on the L2L node so you can determine how close to first gen you want it to be. However it does have the drawback of needing an upscaled version of the first gen to be saved for the L2L node to work. So it produce 3 images in total (the original lower-res version the upscaled L2L driver and the final output)
-
The second (HRO-T2L-RandRange-Param-Example.json) is much simpler and just uses the output of the first generation to drive a tile control net into second T2L at higher resolution. This has the benefit of been quicker than the first method and only producing 2 images (the first gen and final output). In a lot of cases this is a great approach however it can produce images that are vary quite a bit from the original generation, whereas the first method tends to stay much closer. But I have seen a fair few case where the second method produce aesthetically nicer images and also can add more fine detail than the first method. I think both have a place in my toolkit. It should be noted this second method was inspired by the node done by @fiery cloud on his post https://discord.com/channels/1020123559063990373/1130422773496418354
ebon solar
