Create Stunning Asian-Style Portraits with Stable Diffusion: A Workflow Guide
1. Workflow Overview
This workflow generates high-definition images of an ancient Chinese beauty using a Stable Diffusion checkpoint (macc-平面模特_v2.0) and detailed text prompts to create a classical Asian female portrait.
Core Models:
Checkpoint:
macc-平面模特_v2.0(optimized for Asian-style portraits).CLIP Text Encoder: Processes positive/negative prompts.
VAE Decoder: Converts latent images to RGB.
2. Key Components
Nodes:
CheckpointLoaderSimple (Node 4): Loads
macc-平面模特_v2.0, outputs MODEL/CLIP/VAE.CLIPTextEncode (Nodes 6 & 7):
Node 6: Positive prompt (e.g., "ancient Chinese beauty with peach-blossom skin").
Node 7: Negative prompt (e.g., "text, watermark").
EmptyLatentImage (Node 5): Creates blank latent image (784x1136 resolution).
KSampler (Node 3): Uses
Eulersampler, 135 steps, CFG=8.VAEDecode (Node 8): Decodes latent image to final output.
SaveImage (Node 9): Saves image to
ComfyUI/output.
Installation:
All nodes are built-in; no extra installation needed.
Download
macc-平面模特_v2.0.safetensorstomodels/checkpoints.
Dependencies:
Ensure the checkpoint is from CivitAI or official sources.
3. Workflow Structure
Group 1: Model & Prompts
Input: Checkpoint name, positive/negative prompts.
Output: MODEL, CLIP conditioning.
Group 2: Latent Generation
Input: Blank latent (784x1136), sampler settings.
Output: Latent image.
Group 3: Decoding & Saving
Input: Latent image, VAE.
Output: Final image (PNG).
4. Inputs & Outputs
Inputs:
Required:
Positive prompt (detailed beauty description).
Negative prompt (e.g., "text, watermark").
Optional:
Seed (default: random).
Resolution (via
EmptyLatentImage).
Output:
A generated image of an ancient Chinese beauty.
5. Notes
Common Issues:
Missing model: Verify
macc-平面模特_v2.0.safetensorsis installed.GPU OOM: Reduce resolution or steps.
Optimization:
Use
TAESDfor faster previews.Experiment with samplers (e.g.,
DPM++ 2M).