Type | |
Stats | 783 2,957 |
Reviews | (108) |
Published | Aug 26, 2024 |
Base Model | |
Training | Steps: 525 Epochs: 10 |
Usage Tips | Clip Skip: 1 |
Trigger Words | Holographic Cloth glowing fractal patterns in the cloth holographic cloth as if it was drawn by a wire form CAD system showing accurate geometric contours to curved surfaces |
Hash | AutoV2 194372130A |
Made for the Civitai Flux Training Contest
I use the folloring prompt flow for my results for V1. words in orange are the only thing I swap out for training images. The clothes focus tag helps for me:
create a woman wearing holographic (loose t-shirt). gold holographic cloth, depth of field view, clothes focus, home background
[glowing fractal patterns in the cloth] helps with adding the desired texture
This one was a tough one for me, months of planning.
I first had to prove that I could do a proof of concept not for just the glowing effects but also for the digital patterns similar to an old CAD system (TRON)
once I had a prompt that rendered the effects I wanted (70% of the time) I started working with only one piece of clothing at a time with different collors.
After 3 training sessions I'm very impressed on how it turned out :)
the first training images I created was proof of concept:
create a holographic (woman with long hair wearing clothes (t-shirt jeans and sneakers)), white theme, depth of field view, full body focus, simple black background
then I added more verbage:
as if it was drawn by a wire form CAD system showing accurate geometric contours to curved surfaces
Changing from "white theme" to "white holographic cloth" helped with stearing the AI to the piece of clothing I was trying to adjust.
The final result of Training image generation looks like this:
create a woman wearing holographic (loose t-shirt) as if it was drawn by a wire form CAD system showing accurate geometric contours to curved surfaces, violet holographic cloth, depth of field view, clothes focus, [glowing fractal patterns in the cloth]
I use the folloring LoRA's as well:
Training:
V1 - I basicly dragged each image back into ComfyUI and copied and paisted the original prompt into the training for each image. Then I selected the automated prompting. It actually worked out very well for me. (30 images)
V2 - cleaning up training verbage and adding the tag [non-holographic] to help control the LoRA. The original Promts is still used for this version (in process) I'm just tweaking them based on the output, additional tags added like (non-holographic woman) and (non-holographic background) as required. Some images were replaced with better ones and a few were added. (33 images)
TS - Thunder Storm added
I hope this helps