From Brushstrokes to Pixels: A Deep Dive into Stable Diffusion's Graffiti Capabilities

CN
ComfyUI.org
2025-04-29 17:59:29

1. Workflow Overview

ma2dwj8a9n7y2u2co9ra88abb950bd127067a3483a15176db15e981c7c12b5fb59d12965cc63bcd7dd1.png

This workflow specializes in generating "wall graffiti/painting" style images, combining Stable Diffusion with Lora fine-tuning to create Eastern aesthetic murals. Key features include automated Chinese-English prompt translation and Flux diffusion scheduling.

Core Models:

  • BaseAlgo_F.1: Main SD model

  • WallPainting.lora: Style-specific adapter

  • clip_l + t5xxl_fp16: Dual-text encoder

2. Node Breakdown

Critical Nodes:

  1. BaiduTranslateNode

    • Purpose: Auto-translate Chinese prompts to English

    • Install: Via ComfyUI Manager ("BaiduTranslate" required)

    • Note: Requires Baidu Translate API keys

  2. CLIPTextEncodeFlux

    • Purpose: Dual-encoder text processing

    • Dependency: Flux extension package

  3. SamplerCustomAdvanced

    • Config: Euler sampler + simple scheduler

  4. LoraLoaderModelOnly

    • Model Path: /ComfyUI/models/loras/

3. Workflow Structure

Group Analysis:

  • Group 1: Flux Standard

    • Contains: UNETLoader, DualCLIPLoader

  • Group 2: Sampler

    • Key Parameters: 20 steps, random seed

  • Group 3: Prompt/Size

    • Default Resolution: 1440x960

4. I/O Specification

Inputs:

  • Chinese prompt (e.g. "Asian girl with floral patterns")

  • Resolution: 1440x960

  • Lora Strength: 0.8

Outputs:

  • Path: /Flux/ComfyUI/

  • Format: PNG with metadata

5. Critical Notes

  1. Flux extension is mandatory

  2. 8GB+ VRAM recommended for high-res output

  3. Baidu Translate requires API configuration

  4. SDXL base model recommended for best results