This is LoRA that enhances background drawing created experimentally. I collected a large number of background and asset-based LoRAs, balanced them with low weights, merged them, and further strengthened the background using conceptual control. The composition of the final version is not clear, but probably around 130 to 180 LoRAs were merged. Licenses of materials I merged are limited to "No selling models" and "No generation services". I would like to thank all the authors of publicly available LoRAs.
I have tried to accommodate common scenes as much as possible, such as material selection and merging ratio, but it is not perfect. The effect changes depending on the LoRA I collect. This has its strengths and weaknesses, as there are limitations to the materials available. There may be differences in the assets that are likely to appear depending on the compatibility with the applied model. As you increase the ratio, more of the background is shown and the angle of view changes. Please adjust the prompts separately if necessary.
BG Enhance is a collection of many background LoRAs, and naturally it greatly affects the overall picture. In the sample, it is used at around 0.8. 1.0 is usually excessive and often breaks the picture. Therefore, I also prepared a version that uses LoRA Block Weight to have as little effect on the character as possible. Although the effect is slightly weaker, this method can be used more effectively depending on the model. HighRes.Fix and ADtailer are not used in the sample images so that you can check the affected area. Many facial deformities can be corrected.
Description
FAQ
Comments (4)
So no trigger needed? just a .08 wight?
This LoRA is not required trigger word. However, there may be a possibility that the trigger word set in the merged LoRA will respond. Since it is a merge with a very low weight, there was almost no response as far as I tried it with one LoRA-specific trigger word, but it is possible if it is a more general concept.
As for the weight when using it, this time I tried all with 0.8. But, 0.8 may be a bit excessive.
Please see my Image post. As long as it is used normally at around 0.5, there will be no reaction even if a specific trigger word is used, and the effect of BG Enhance is only to improve the texture of stones etc. The specific merged LoRA will not react. If you increase the weight of LoRA beyond the limit, you can confirm that it is effective to the extent that it is not completely meaningless.
Huh, this is interesting. In what I've seen so far, it's barely distinguishing between foreground and background, and the foreground details it creates are undesirable. But when used as a negative, in conjunction with "more_details" lora, the two balance each other out really really nicely.