Local Image Generation Setup (Built-in Engine)
DREAMIO's built-in engine allows you to run image generation models directly on your own hardware. This guide will walk you through the setup process and explain the various settings available.
Step 1: Download an Image Model
To generate images locally, you'll first need to download a model file. These models are typically in formats like .safetensors or .ckpt.
- Where to find models: A popular resource for image models is Civitai. You can also find them on Hugging Face.
- Choosing a model: Look for models that match the art style you want to create. Pay attention to the model's base version (e.g., SD1.5, SDXL) as this can affect performance and compatibility.
- Organize your files: It's a good practice to create a dedicated folder on your computer to store your image models.
Step 2: Configure DREAMIO
Once you have a model, you need to tell DREAMIO where to find it and how to use it.
- Launch DREAMIO and navigate to Image Generation Settings.
- Set the Provider to
Local.
This will reveal the following settings:
Core Settings
- Models Path: Click to select the folder on your computer where you saved your image model files.
- Model: Once the path is set, choose the specific model file (e.g.,
your-model.safetensors) from this dropdown menu. - Keep in Memory: Keep the model loaded in your computer's RAM/VRAM for faster generation. This consumes significant resources, so only enable it if you have enough memory.
Generation Parameters
These settings control the quality and style of the final image.
- Dimensions: The resolution of the generated image (e.g., 512x512, 1024x1024). Higher resolutions require more VRAM.
- Sample Steps: The number of iterations the AI takes to generate an image. More steps can improve quality but take longer.
- Sampler: The algorithm used to generate the image. Different samplers can produce stylistically different results.
- Cfg Scale: How strictly the AI should follow your visual description. Lower values give the AI more creative freedom, while higher values make it adhere more closely to the prompt.
- Scheduler: The noise scheduler used during image generation.
Advanced Settings
These settings are for users who want more fine-grained control over the generation process.
- Backend: The processing engine to use. Select
CUDAfor NVIDIA GPUs,Vulkanfor AMD/Intel GPUs, orCPUto use your main processor (slowest). - Quantization: The data format for the model's weights (e.g.,
Q8_0,F16,F32). - TAESD Path/Model: You can specify a "tiny autoencoder" to speed up the final stage of image decoding. This is an optional optimization.
- LoRA Path/LoRA: Settings for LoRA (Low-Rank Adaptation) models. These are small files that can be used to modify the style or add specific concepts to your main model.
- VAE Tiling: A memory-saving option that can help prevent "out of memory" errors on GPUs with less VRAM.
- UNET/CLIP/VAE Paths & Models: For ultimate control, you can specify individual file paths for the core components of the diffusion pipeline: the UNET, CLIP model, and VAE. In most cases, these are bundled within the main model file and don't need to be set manually.
Alternative: Using ComfyUI
For users who want maximum flexibility and control over the image generation process, DREAMIO also supports integration with ComfyUI.
ComfyUI is a powerful and modular node-based interface for Stable Diffusion. It allows you to build complex workflows, chain models together, and experiment with advanced techniques that go beyond the capabilities of the built-in engine.
If you are an advanced user or want to leverage a specific ComfyUI workflow, you can connect DREAMIO to your running ComfyUI instance.