MOSS Image - Color-Depth-Control LoRA
Control adapter trained by darask0 for MOSS Image.
简介
This ComfyUI workflow demonstrates how to guide MOSS Image generation with a Color-Depth Control LoRA trained by darask0. It lets you steer structure with a depth map while preserving or modulating color from a reference image, making it easy to create style variations that keep geometry intact. At its core, LoadImage brings in your RGB reference and (optionally) a precomputed depth map, while PreviewImage and ImageCompare help you iterate quickly and check differences across runs. SaveImageAdvanced handles clean, configurable exports.
你需要提供
- 正在读取工作流
在 ComfyUI 中运行。 此工作流无法在本页面运行。你可以在 Comfy Cloud 中打开（已预先加载所需内容），或下载后在自己的 ComfyUI 中使用。
在 Cloud 中试用常见问答
The workflow is designed to leverage a depth map for structural guidance, which is strongly recommended for best results. If you don’t have one, generate it first with a depth-preprocessor node (e.g., MiDaS/Depth Anything) or an external tool, then load it alongside your color image.
You’ll need the MOSS Image base model and the Color-Depth Control LoRA trained by darask0. Place them in your ComfyUI model directories where your installation expects checkpoints/LoRAs, then select them inside the custom MOSS nodes (IDs 3d58c0bd-7a87-40a7-bdb5-a209a27f10d7 and 882218ff-0ea7-4047-9b12-b79cdc4e9faa). Reload models if they don’t appear.
Ensure the RGB image and depth map share the same resolution and alignment. Use standard grayscale depth maps (near=brighter, far=darker) and avoid extreme upscaling. If you still see artifacts, lower the adapter/control strength, reduce output resolution, and keep a fixed seed while you iterate.
This workflow focuses on image-guided control (color+depth). For pure text-to-image with MOSS, use a text-to-image workflow built for that purpose. You can still combine prompts with this pipeline if the custom node exposes text inputs, but depth guidance remains the main driver.