2 ms·
IIRC it does merging/weighting behind the scenes.
by minimaxir 3y ago
IIRC it does merging/weighting behind the scenes.
- Der_Einzige 3y agoSource for this?
- cheald 3y agoI'm pretty sure that it's just serially summing the network weights, which results in an accumulated offset to the self-attention layers of the transformer. It's not doing any kind of analysis of multiple networks prior to application to make them "play nice" together; it's just looping and summing. https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/master/extensions-builtin/Lora/networks.py#L285-L313 https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob...