2d-art-sprites Dreambooth model trained by ana-tamais with TheLastBen's fast-DreamBooth notebook
You can test this model using this Colab Notebook for Inference
Sample pictures of 2dart concept:
We saved the training data in dataset.zip, and some generated results in results.zip.
Some recommendations
We recommend to set the:
- prompt as:
"[some wizard, paladin, healer, etc.], in the style of 2dart, white background, no background, full body" - negative prompt as:
"deformed, mutilated limbs, background, multiple people" - guidance scale between
9 and 10 - sampling method as
Euler - sampling steps as
60 - batch size as
1- avoid high batch size, unless you have a high memory GPU
This set of hyperparameters lead us to stable and good results.
This model was trained using images which the character has some human format or is a not deformed living being. So if you try to predict something like "sword, mirror, candle, etc" (non-living things), we saw the model doesn't perform so well.
You need at least a Tesla T4 to be able to run the inference step using the given notebook.
- Downloads last month
- 25
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.