Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,53 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Model Card for RoboDiffusionXL: Advanced Robotic Imagery LORA Model
|
2 |
+
|
3 |
+
## Setup and Initialization
|
4 |
+
|
5 |
+
This model must not be used at full strength but at approximately 70%. E.g. in Auto1111 Forge... <lora:robodiffusionxl:0.7>
|
6 |
+
|
7 |
+
## The main keywords for this model are:
|
8 |
+
- Quadruped
|
9 |
+
- Hexapod
|
10 |
+
- Octopod
|
11 |
+
- Centiped
|
12 |
+
- Aerial
|
13 |
+
- Wheeled
|
14 |
+
- Underwater
|
15 |
+
|
16 |
+
Choose the appropriate keyword type for the desired motion type for the robot.
|
17 |
+
|
18 |
+
## Model Details
|
19 |
+
|
20 |
+
- **Model Name:** RoboDiffusionXL
|
21 |
+
- **Version:** 1.0
|
22 |
+
- **Model Type:** Image Generative LORA Model based on SDXL Base
|
23 |
+
- **Developers:** Fiacre
|
24 |
+
- **Release Date:** May 20, 2024
|
25 |
+
- **Model Repository:** [Hugging Face Models Hub](https://huggingface.co/Fiacre/robodiffusion-xl-v1)
|
26 |
+
|
27 |
+
## Overview
|
28 |
+
|
29 |
+
RoboDiffusionXL is a LORA (Latent Optimization with Representational Adjustment) based on the SDXL (Stable Diffusion XL) architecture. It is specially designed for generating high-quality, diverse images of robots in various forms, including but not limited to wheeled, quadruped, hexapod, octopod, centipede, underwater, and aerial robots, across multiple artistic styles.
|
30 |
+
|
31 |
+
## Training Data
|
32 |
+
|
33 |
+
RoboDiffusionXL was trained on a high-quality synthetic dataset curated to include a wide variety of robotic forms and styles. The images include historical, cultural, and futuristic themes, ensuring diverse generated outputs.
|
34 |
+
|
35 |
+
## Key Configuration and Settings
|
36 |
+
|
37 |
+
- **Optimizer:** Adafactor.
|
38 |
+
- **Learning Rate:** 0.0009.
|
39 |
+
- **Batch Size:** 5.
|
40 |
+
- **Rank:** 256 (not so low rank), but was required otherwise the image were poor.
|
41 |
+
|
42 |
+
## Limitations
|
43 |
+
|
44 |
+
- Limited styles.
|
45 |
+
- It cannot do triped, and quintaped robots well.
|
46 |
+
|
47 |
+
## Licensing and Usage
|
48 |
+
|
49 |
+
license: openrail
|
50 |
+
|
51 |
+
## Future Work
|
52 |
+
|
53 |
+
Future updates will include the introduction of triped and quintaped robots, alongside a broader array of diverse styles. The aim is to continuously expand the model's capabilities to cover an even wider spectrum of robotic forms and artistic interpretations. Community suggestions are appreciated.
|