Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,40 @@
|
|
1 |
---
|
2 |
license: openrail++
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: openrail++
|
3 |
---
|
4 |
+
|
5 |
+
## Description
|
6 |
+
|
7 |
+
Maxwell the Cat Diffusion is a latent text-to-image diffusion model based on the original CompVis Stable Diffusion v1.5 and then fine-tuned on 5 images based on the 'Maxwell the Cat' meme originating from the modded Half Life 2 video.
|
8 |
+
|
9 |
+
To use this gorgeous object in your generations, add `maxwell the cat` to the prompts.
|
10 |
+
|
11 |
+
## Dreambooth hyperparameters
|
12 |
+
|
13 |
+
```sh
|
14 |
+
export MODEL_NAME="runwayml/stable-diffusion-v1-5"
|
15 |
+
export INSTANCE_DIR="/home/kabachuha/kml/datasets/objects/maxwell"
|
16 |
+
export CLASS_DIR="/home/kabachuha/kml/datasets/objects/maxwell_class"
|
17 |
+
export OUTPUT_DIR="/home/kabachuha/kml/maxwell/"
|
18 |
+
|
19 |
+
accelerate launch train_dreambooth.py \
|
20 |
+
--pretrained_model_name_or_path=$MODEL_NAME \
|
21 |
+
--instance_data_dir=$INSTANCE_DIR \
|
22 |
+
--class_data_dir=$CLASS_DIR \
|
23 |
+
--output_dir=$OUTPUT_DIR \
|
24 |
+
--with_prior_preservation --prior_loss_weight=1.0 \
|
25 |
+
--instance_prompt="maxwell the cat" \
|
26 |
+
--class_prompt="3d model of a black cat, lowpoly" \
|
27 |
+
--resolution=512 \
|
28 |
+
--train_batch_size=1 \
|
29 |
+
--gradient_accumulation_steps=1 \
|
30 |
+
--learning_rate=1e-6 \
|
31 |
+
--lr_scheduler="constant" \
|
32 |
+
--lr_warmup_steps=0 \
|
33 |
+
--num_class_images=200 \
|
34 |
+
--max_train_steps=800 \
|
35 |
+
--mixed_precision 'no' \
|
36 |
+
--train_text_encoder \
|
37 |
+
--checkpointing_steps 1200
|
38 |
+
```
|
39 |
+
|
40 |
+
The dataset link https://drive.google.com/drive/folders/1nd8NHrwuu_VKHaU8iBFw95w1iqfmhNTo?usp=share_link
|