This isn’t about a bug, but rather a proposal for an official node in ComfyUI.

[DaveGravel]

I've noticed that many users enjoy modifying and merging models with LoRAs. It’s great that this is possible, but often, some try to present these merges as entirely new models, as if they were the original creators. The same goes for the LoRAs used in the merges, which they often don't even own.

My idea is to create a basic node that can save some model configuration settings, for example, the base model like Flux, as well as the LoRAs used and their strengths.

This way, we would only need to keep the original Flux model, which saves disk space and a lot of internet bandwidth. The node would load a configuration file containing both the inference settings and the LoRAs to apply.

With this setup, only the LoRAs would need to be downloaded. The original model would remain unchanged, and we wouldn't need to store 10 different versions of the same model just because each one is merged with different LoRAs. That would be a huge space saver.

Plus, the configuration would remain editable, unlike merges, which are not easily modifiable afterward.

I know it doesn't cover all types of merges or model modifications, some changes go beyond just LoRAs, but it can help reduce the number of incorrect or misleading models circulating online.

[Kush210]

Totally agreed with the way Lora should be use or released, just download the LoRA instead of the merged lora

[ltdrdata]

If hash management for models becomes more established, organizing them in the form of recipe files based on those hashes could be a good idea.