ComfyKit Custom Nodes
Nodes:LoraWithMetadata, TypecasterImage.
Nodes (9)
LoRA S5 in practice
The retired three-LoRA stack you'll still find in old workflows
Three LoRAs, split strengths, and the .tpNN trigger convention nobody documented
A cast for the text-encoder wire
Recast Conditioning
The no-op that fixes broken links
A cast for the compressed pixel space
Forcing a type on the model wire
A typed pass-through for your decoder
ComfyKit Custom Nodes
LoraStackFiveSimple (LoRA S5)
Accepts 5 LoRAs to blend and a weight value to apply. It applies the same weight to both the Model and CLIP as I never found myself with an obvious positive result when desyncing them. Does not mean it cannot, I just do not and therefore did this to save screen space on the home view of the workflow.
Typecaster
(CLIP, Conditioning, Image, Latent, Model, VAE)
Recasts the input as typed output without doing anything to it. This helps with workflows where one node can output "anything" and send it to a node that can take "anything". These type of nodes often struggle the first time a workflow is loaded in the browser as their wiring accepts anything but may not notice what kind of anything it had until you queue a job once and let it fail.
In code it would be the same as being like emptyLatent = (Latent)emptyLatent;