prithivMLmods
commited on
Commit
•
cc3d3e0
1
Parent(s):
fd0b5af
Update README.md
Browse files
README.md
CHANGED
@@ -24,8 +24,47 @@ license: creativeml-openrail-m
|
|
24 |
![strangerzonehf/Flux-3DXL-Partfile-0004](images/4x.png)
|
25 |
|
26 |
<Gallery />
|
|
|
27 |
|
|
|
28 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
## Trigger words
|
30 |
|
31 |
You should use `3DXLP4` to trigger the image generation.
|
|
|
24 |
![strangerzonehf/Flux-3DXL-Partfile-0004](images/4x.png)
|
25 |
|
26 |
<Gallery />
|
27 |
+
## Model description for 3DXL Partfile 0004
|
28 |
|
29 |
+
Image Processing Parameters
|
30 |
|
31 |
+
| Parameter | Value | Parameter | Value |
|
32 |
+
|---------------------------|--------|---------------------------|--------|
|
33 |
+
| LR Scheduler | constant | Noise Offset | 0.03 |
|
34 |
+
| Optimizer | AdamW | Multires Noise Discount | 0.1 |
|
35 |
+
| Network Dim | 64 | Multires Noise Iterations | 10 |
|
36 |
+
| Network Alpha | 32 | Repeat & Steps | 17 & 2550 |
|
37 |
+
| Epoch | 20 | Save Every N Epochs | 1 |
|
38 |
+
|
39 |
+
Labeling: florence2-en(natural language & English)
|
40 |
+
|
41 |
+
Total Images Used for Training : 14
|
42 |
+
|
43 |
+
## Best Dimensions & Inference
|
44 |
+
|
45 |
+
| **Dimensions** | **Aspect Ratio** | **Recommendation** |
|
46 |
+
|-----------------|------------------|---------------------------|
|
47 |
+
| 1280 x 832 | 3:2 | Best |
|
48 |
+
| 1024 x 1024 | 1:1 | Default |
|
49 |
+
|
50 |
+
### Inference Range
|
51 |
+
- **Recommended Inference Steps:** 30–35
|
52 |
+
|
53 |
+
## Setting Up
|
54 |
+
```python
|
55 |
+
import torch
|
56 |
+
from pipelines import DiffusionPipeline
|
57 |
+
|
58 |
+
base_model = "black-forest-labs/FLUX.1-dev"
|
59 |
+
pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=torch.bfloat16)
|
60 |
+
|
61 |
+
lora_repo = "strangerzonehf/Flux-3DXL-Partfile-0004"
|
62 |
+
trigger_word = "3DXLP4"
|
63 |
+
pipe.load_lora_weights(lora_repo)
|
64 |
+
|
65 |
+
device = torch.device("cuda")
|
66 |
+
pipe.to(device)
|
67 |
+
```
|
68 |
## Trigger words
|
69 |
|
70 |
You should use `3DXLP4` to trigger the image generation.
|