Sign In

💔 Split Heart Necklace SDXL 💔 - Custom broken heart necklace

15
52
6
5
Updated: Aug 23, 2024
conceptheartnecklacebroken
Verified:
SafeTensor
Type
LoRA
Stats
52
6
Reviews
Published
Aug 10, 2024
Base Model
SDXL 1.0
Usage Tips
Strength: 1
Trigger Words
SplitHeart, MaterialHeart heart, LeftSide|RightSide, MaterialChain chain, BackgroundColor background
Training Images
Download
Hash
AutoV2
4D63829F40
Text-Tacular Contest Winner
MN
mnemic

💔 Have some heartache to share? 💔

💔 Did you drop one of your hobbies? 💔

This is the right LoRA for you!

This model lets you (in theory) create a broken heart necklace out of any 2 subjects. One for the left side, and one for the right. The prompting is strict, so please follow the guide below.

Prompting / Usage

You "need "to use the following formula when using this model:

SplitHeart, MaterialHeart heart, LeftSide|RightSide, MaterialChain chain, BackgroundColor background

Let's break this down

SplitHeart, : This is the trigger word for the LoRA. I suggest using the comma and space after as well.
MaterialHeart: This should control the material of the heart itself. See the Tips-section below for some material suggestions.
heart: This should be the shape of the object. It should be a heart, but you may get success with other shapes as well. I haven't experimented a lot with this part yet.
Leftside|Rightside: This is what controls the subjects of the two different heart halves. It should be separated with a | character (not a l or a 1).

Sometimes it works perfectly, but it's unfortunately still common to get concept bleeding, meaning what you type on one side can affect the other side.

See the Tips-section for more suggestions to control this.
MaterialChain chain: This is the material of the chain. It's trained on a couple of different ones. See the Tips-section below for suggestions.
BackgroundColor background: The model is mostly trained on simple backgrounds. They are all captioned so you should be able to change it by prompting it with a color and describing your background. See the example images for suggestions.

The model may work without this prompting structure, but this is how it was trained and how I get good results.

The model works best with square image ratios, but can also work in others.

Note:

I may be using the following extensions in the reference images: FreeU or ADetailer / Face Editor. I may also be using a specific Hires Checkpoint (swaps to another model when doing the hires steps). If your results don't match mine exactly, this may explain why.

Tips and tricks

Below are some tips and tricks for how to get even more out of the model.

Materials List

Try using some of the materials from the list below as materials for your heart:

Marble
Iron
Ivory
Wood
Crystal
Iron
Silver
Gold
Plastic
Cloth
Paper
Ceramic
Graphite
Glass

Chain Materials List

Try using some of the materials from the list below as suggestions for the chain:

Black Metallic chain
Yellow Coord chain
Rainbow Metallic chain
Glitter Gold chain
Opalescent Pearl chain
Brown Leather Strap chain
Paper chain
Titanium Metal chain
Gold Shiny chain

Improving Subjects

If you are not getting the results you need, it's not unlikely there is some concept bleeding. Here are a few ideas and suggestions to get different results and improve it.

Use Inpainting
First generate a heart, and then mask out the different areas you want and inpaint the right object inside. Not great but it could help you get there
Use BREAK
Using BREAK inside your prompt may help you get more consistent results. Try adding a BREAK after the first subject.
Use Attention Masking / Regional Prompting
https://github.com/hako-mikan/sd-webui-regional-prompter
Something like this could help you with different prompts for different parts of the image. I haven't tried it, but I suspect it will work well for this model.

Link to Training Data

Like it?

https://ko-fi.com/mnemic

https://www.buymeacoffee.com/mnemic