prithivMLmods
commited on
Commit
•
ab5d3fd
1
Parent(s):
0522c25
Update README.md
Browse files
README.md
CHANGED
@@ -39,11 +39,51 @@ license: openrail
|
|
39 |
![Ghibli.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/Eq6JzdtLuwJhRINHSNRgt.png)
|
40 |
|
41 |
# Flux-Ghibli-Art-LoRA
|
|
|
42 |
<Gallery />
|
43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
|
|
|
|
|
45 |
|
|
|
|
|
|
|
46 |
|
|
|
|
|
|
|
47 |
## Trigger words
|
48 |
|
49 |
You should use `Ghibli Art` to trigger the image generation.
|
|
|
39 |
![Ghibli.png](https://cdn-uploads.huggingface.co/production/uploads/65bb837dbfb878f46c77de4c/Eq6JzdtLuwJhRINHSNRgt.png)
|
40 |
|
41 |
# Flux-Ghibli-Art-LoRA
|
42 |
+
|
43 |
<Gallery />
|
44 |
|
45 |
+
## Model description for Flux-Ghibli-Art-LoRA
|
46 |
+
|
47 |
+
Image Processing Parameters
|
48 |
+
|
49 |
+
| Parameter | Value | Parameter | Value |
|
50 |
+
|---------------------------|--------|---------------------------|--------|
|
51 |
+
| LR Scheduler | constant | Noise Offset | 0.03 |
|
52 |
+
| Optimizer | AdamW | Multires Noise Discount | 0.1 |
|
53 |
+
| Network Dim | 64 | Multires Noise Iterations | 10 |
|
54 |
+
| Network Alpha | 32 | Repeat & Steps | 15 & 2400 |
|
55 |
+
| Epoch | 15 | Save Every N Epochs | 1 |
|
56 |
+
|
57 |
+
Labeling: florence2-en(natural language & English)
|
58 |
+
|
59 |
+
Total Images Used for Training : 20
|
60 |
+
|
61 |
+
## Best Dimensions & Inference
|
62 |
+
|
63 |
+
| **Dimensions** | **Aspect Ratio** | **Recommendation** |
|
64 |
+
|-----------------|------------------|---------------------------|
|
65 |
+
| 1280 x 832 | 3:2 | Best |
|
66 |
+
| 1024 x 1024 | 1:1 | Default |
|
67 |
+
|
68 |
+
### Inference Range
|
69 |
+
|
70 |
+
- **Recommended Inference Steps:** 30–35
|
71 |
+
|
72 |
+
## Setting Up
|
73 |
+
```python
|
74 |
+
import torch
|
75 |
+
from pipelines import DiffusionPipeline
|
76 |
|
77 |
+
base_model = "black-forest-labs/FLUX.1-dev"
|
78 |
+
pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=torch.bfloat16)
|
79 |
|
80 |
+
lora_repo = "strangerzonehf/Flux-Ghibli-Art-LoRA"
|
81 |
+
trigger_word = "Ghibli Art"
|
82 |
+
pipe.load_lora_weights(lora_repo)
|
83 |
|
84 |
+
device = torch.device("cuda")
|
85 |
+
pipe.to(device)
|
86 |
+
```
|
87 |
## Trigger words
|
88 |
|
89 |
You should use `Ghibli Art` to trigger the image generation.
|