laloadrianmorales commited on
Commit
be292cf
·
verified ·
1 Parent(s): b6cb46d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +47 -71
README.md CHANGED
@@ -1,84 +1,60 @@
1
- # Matrix-Game-2.0 Hugging Face Space Setup
2
-
3
- ## 🚀 Quick Setup Guide
4
-
5
- ### 1. **Create the Space**
6
- - Go to [Hugging Face Spaces](https://huggingface.co/spaces)
7
- - Click "Create new Space"
8
- - Choose **Gradio** as the SDK
9
- - **IMPORTANT**: Set Hardware to **GPU (A100)** - this is crucial!
10
-
11
- ### 2. **Upload Files**
12
- Upload these three files to your Space:
13
-
14
- 1. **app.py** - The main application (use the clean version provided)
15
- 2. **requirements.txt** - Dependencies list (properly formatted)
16
- 3. **README.md** - This file (optional)
17
-
18
- ### 3. **Space Configuration**
19
- In your Space settings, ensure:
20
- - **SDK**: Gradio
21
- - **Hardware**: A100 GPU (required!)
22
- - **Visibility**: Public or Private (your choice)
23
-
24
- ### 4. **Key Improvements Made**
25
-
26
- #### ✅ **Fixed Issues:**
27
- - Proper requirements.txt formatting (each dependency on new line)
28
- - Added missing dependencies (opencv, xformers, etc.)
29
- - Used `@spaces.GPU` decorator for proper GPU allocation
30
- - Simplified model downloading and setup
31
- - Added proper error handling and timeouts
32
- - Reduced frame limits for HF Spaces constraints
33
-
34
- #### ✅ **HF Spaces Optimizations:**
35
- - Pre-download model weights efficiently
36
- - Use temporary directories properly
37
- - Clean up resources after generation
38
- - Progress indicators for user feedback
39
- - GPU memory management
40
 
41
- ### 5. **Usage Tips**
42
- - **First run**: Will take 2-3 minutes to download model (~10GB)
43
- - **Subsequent runs**: Much faster (model cached)
44
- - **Recommended**: 25-75 frames for reasonable generation time
45
- - **Input images**: Works best with landscapes/scenes, max 512px
46
 
47
- ### 6. **Troubleshooting**
48
 
49
- #### If you get "CUDA out of memory":
50
- - Reduce number of frames (try 25-50)
51
- - Use smaller input images
52
- - Restart the Space
53
 
54
- #### If model download fails:
55
- - Check internet connection in Space logs
56
- - Verify you're using A100 GPU hardware
57
- - Try restarting the Space
 
58
 
59
- #### If generation is slow:
60
- - This is normal! Video generation is compute-intensive
61
- - A100 GPU should handle 50 frames in ~2-3 minutes
62
- - Consider reducing frames for faster results
63
 
64
- ### 7. **Expected Performance**
65
- - **Model download**: 2-3 minutes (first time only)
66
- - **25 frames**: ~1 minute generation
67
- - **50 frames**: ~2-3 minutes generation
68
- - **75+ frames**: 4-5+ minutes generation
69
 
70
- ---
71
 
72
- ## 🎮 What is Matrix-Game-2.0?
 
 
 
 
73
 
74
- Matrix-Game-2.0 is an interactive world foundation model that generates real-time video content based on input images. It uses advanced diffusion techniques to create dynamic, controllable video sequences that respond to user interactions.
75
 
76
- **Key Features:**
77
- - Real-time video generation (up to 25 FPS)
78
- - Interactive control through keyboard/mouse inputs
79
- - Long video generation capability
80
- - Fine-grained scene control and evolution
81
 
82
  ---
83
 
84
- *Built with ❤️ for the Hugging Face community*
 
1
+ ---
2
+ title: Matrix-Game-2.0
3
+ emoji: 🎮
4
+ colorFrom: purple
5
+ colorTo: blue
6
+ sdk: gradio
7
+ sdk_version: 4.42.0
8
+ app_file: app.py
9
+ suggested_hardware: a100-large
10
+ suggested_storage: large
11
+ python_version: 3.10
12
+ short_description: Interactive World Model for Real-Time Video Generation
13
+ tags:
14
+ - video-generation
15
+ - diffusion
16
+ - interactive
17
+ - world-model
18
+ - matrix-game
19
+ models:
20
+ - Skywork/Matrix-Game-2.0
21
+ pinned: false
22
+ ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
23
 
24
+ # 🎮 Matrix-Game-2.0
 
 
 
 
25
 
26
+ An interactive world foundation model for real-time video generation. Upload an image and generate dynamic video content with controllable parameters!
27
 
28
+ ## 🚀 How to Use
 
 
 
29
 
30
+ 1. **Upload an Image**: Choose a landscape or scene image (works best)
31
+ 2. **Set Parameters**:
32
+ - **Frames**: 25-100 (lower = faster generation)
33
+ - **Seed**: Any number for reproducibility
34
+ 3. **Generate**: Click the button and wait 2-5 minutes
35
 
36
+ ## 💡 Tips
 
 
 
37
 
38
+ - **First run** takes longer (downloads model weights ~10GB)
39
+ - **Best inputs**: Clear, well-lit outdoor scenes or landscapes
40
+ - **Frame recommendations**: 25-50 frames for faster results
41
+ - **Hardware**: Requires GPU acceleration (A100 recommended)
 
42
 
43
+ ## ⚙️ Technical Details
44
 
45
+ Built on the Matrix-Game-2.0 model by SkyworkAI, this Space provides:
46
+ - Real-time video generation up to 25 FPS
47
+ - Interactive control through diffusion-based synthesis
48
+ - Long video generation capabilities
49
+ - Fine-grained scene evolution
50
 
51
+ ## 🔧 Model Information
52
 
53
+ - **Base Model**: Skywork/Matrix-Game-2.0
54
+ - **Architecture**: Auto-regressive diffusion
55
+ - **Input**: Single image + parameters
56
+ - **Output**: MP4 video (configurable length)
 
57
 
58
  ---
59
 
60
+ *Powered by Hugging Face Spaces 🤗*