Makes Christmas lights glow brighter and kicks the festive atmosphere up a notch.
Version 1.0 is really quite over the top, it will add Christmas lights everywhere if it can. Because all the training images are outdoor, you may need to prompt a bit harder for indoor scenes. If you find the effect too strong or the prompt is not being followed, turn down the weight and try again 😅. See this post for comparison of the LoRA at strength 1.0, 0.9 and 0.7.
The alpha 0.5 version needs more baking, but it is fun to play with already. Note that the 2nd Grumpy Cat image was generated without using the LoRa, so that you can see its effect.
Have fun, and if you post your first image directly here (click on "Add Post" above the gallery), I will give you 10 buzz for that image 😎 (If I forgot, just complain loudly 😹)
Description
The training set consists mostly of images from The Most Outrageous Christmas Light Displays of All Time.
Clean up the dataset by removing blurry images. Split up 640x480 images into two 512x480 so that there is no down scaling.
This version is really quite over the top, putting Christmas light wherever there is room to put one in. Because the training set are all outdoor shots, you need to prompt a bit harder for indoor scenes.
This is epoch 11, the rest can be found here: tensor. art/models/808704872812670351/Christmas-Lights-3-NoCD6Co5E11-2024-12-18-21:47:39-Ep-11
FLUX.1 - dev-fp8
Trigger: christmas lights
Repeat: 20 Epoch: 11 (Seventeen 512x512 image total steps: 4080)
Unet LR: 0.0005 Scheduler: cosine Optimizer: AdamW
Network Dim: 6 Alpha: 3
Epoch Loss
1 0.425
2 0.427 <-
3 0.414
4 0.398
5 0.386
6 0.377
7 0.376
8 0.256
9 0.253
10 0.242 <- lowest
11 0.252