PixelOn is an extension that adds On-Device image generation capabilities to Piskel, the PixelArt Editor. This allows users to create pixel art using AI image generation easily and quickly, anywhere and anytime.
Boost your expressiveness!
We used Stable Diffusion C++ to implement lightweight and fast image generation (4-5 seconds on NVIDIA RTX 3070). Combined with Piskel, it provides a seamless creative experience.
This project was designed and developed to allow anyone to use Generative AI on their own computer as GPU adoption in households increases, moving away from cloud-based generators that require continuous subscription fees.
- AI Image Generation: Generate pixel art via text prompts
- Compatibility: Fully compatible with existing
.piskelfiles - Export: Immediately export generated images to Editor layers
| Type | Minimum | Recommended |
|---|---|---|
| CPU | Intel i5 or higher | Intel i5 or higher |
| RAM | 16GB or higher | 16GB or higher |
| GPU | NVIDIA RTX 2000 Series (4GB VRAM+) | NVIDIA RTX 3070 or higher |
| Disk | 20GB+ Free Space | 20GB+ Free Space |
Notice 1: CPU-only usage is possible, but it may take 2-3 minutes or more per image. (Approx. 10s on RTX 2060) Notice 2: AMD GPU support will be updated in the future.
- Stable Diffusion C++
- Piskel
- Stable Diffusion 1.5 Models & LoRAs
- For information about the AI models and LoRAs used, please refer to the AI Models Used section below.
- Click Download PixelOn Installer to get the installation file.
- Run the installer to automatically scan your system and select the optimal acceleration version (CUDA, etc.).
- Once installation is complete, run PixelOn from your desktop.
- Q. I have an Nvidia GPU but CUDA is not selected.
- A. Your GPU driver might be outdated. Please update to the latest version at Nvidia Driver Download.
- Q. I have an AMD GPU, but ROCm is not selected or errors occur during execution.
- A. Your GPU driver might be outdated. Please update to the latest version at AMD Drivers Support.
- A. Additionally, ROCm support is currently experimental due to limited testing environments. Please let us know if you encounter any issues!
- Q. It says "Server Connection Failed".
- A. There may be missing files due to network issues during installation. Please reinstall via the installer.
- Q. Generation speed is too slow.
- A. Speed can be slow if you don't have a GPU accelerator or meet the specs. This is a physical limitation of local execution.
- Q. How do I uninstall PixelOn?
- A. You can uninstall it by running PixelOn Installer.exe again, or by searching for PixelOn in Windows
Settings > Apps > Installed apps.
- A. You can uninstall it by running PixelOn Installer.exe again, or by searching for PixelOn in Windows
When you run PixelOn, the Piskel editor screen appears. You can perform dot work just like existing Piskel usage. For detailed usage, please refer to the Piskel Official Guide.
You can quickly generate images using the prompt window at the bottom of the editor. Only Positive Prompts can be entered in this window.
Open the Detail window for more precise work. Detailed descriptions of each component are below.
|
Manage previous work sessions in the History tab.
|
|
Manipulate key settings for image generation.
Note: If the resolution is too small, the image may not generate correctly. |
|
Check and manage generated images.
Warning: Deletion cannot be undone. |
|
Check logs or send images to Editor.
|
PixelOn uses multiple tags to compose prompts. After entering text, press the Enter key to register a tag. The entered tags are internally connected and used for image generation.
- Don't make it too complex: Each tag should be written as a noun or adjective + noun, preferably within 3-4 words.
- Set appropriate resolution: Resolutions smaller than 64x64 make it difficult to express details.
- For General, Character, SD Character models, 64x64 is recommended.
- For Background models, 128x128 is recommended.
- Try multiple times: AI generation is probabilistic, so the same prompt can produce different results each time.
- Tag order matters: Tags positioned earlier have a greater influence on image generation.
- Use Negative Prompts: Specifying unwanted elements (e.g., garish, amateur) improves quality.
Preset: General |
Preset: Character |
Preset: SD Character |
Preset: Background |
PixelOn utilizes the following open-source AI models:
- Base Model (Stable Diffusion 1.5): Cetus-Mix, QteaMix
- LoRA: 8bitdiffuser 64x, pixel world, pixelartredmond-1-5v-pixel-art-loras-for-sd-1-5
Each model follows its respective license. Please check each model's license for commercial use.
Each preset uses the following model combinations:
- General: Cetus-Mix base + pixelartredmond-1-5v-pixel-art-loras-for-sd-1-5 LoRA
- Character: Cetus-Mix base + 8bitdiffuser 64x LoRA
- SD Character: QteaMix base + 8bitdiffuser 64x LoRA
- Background: Cetus-Mix base + pixel world LoRA
- Prepare an environment that meets the above prerequisites.
- Run
bootstrap.batto install necessary dependencies.- Includes: Stable Diffusion C++, Piskel (Forked Repo), AI Models
- Run
python app.pyto start the local server.
Please use the dev branch if you want to contribute. Pull Requests are always welcome!
This project follows the MIT License.