Type | |
Stats | 99 541 120 |
Reviews | (17) |
Published | Apr 26, 2025 |
Base Model | |
Training | Steps: 1,160 Epochs: 10 |
Usage Tips | Strength: 1 |
Trigger Words | GI-TCG |
Hash | AutoV2 292F5098D8 |
I know there is already a LoRA of this style on Civitai, but it has not been updated and it needs some variety. This was trained on 90 playable official TCG cards at the time I made this (I think that was 4.7, so June 2024), however I plan on updating this when new ones are added. As for how well it works, it is kind of mixed. I didn't really curate the tags at all, so there are a ton of random tags in there. The eyes and fingers are often super bad (which I know is normal for ai). You may want to add things like matte hair shading, thick outlines and colored outlines to your prompts. The outlines may be in weird places and proportions are also not always correct. It has a lot of trouble with character LoRAs, the style will either just not be correct or it will be overbaked and the image will look deep-fried. The style is also just not very strong and it can be overridden by other checkpoints and LoRAs easily. Other than all that, I'd say it captures the style of the cards pretty well.
Pony and Illustrious versions are definitely much better than the SD1.5 version. The Illustrious version is a little weak, so you may want the weight above 1 (see second image).