English

← Back to list

Windows 11 installation ComfyUI and Stable Diffusion production of text drawings - feeding tutorial

itadol5j

I. Environmental preparedness

Hardware requirements

  • graphic cards: NVIDIA graphic card (recommended RTX 30/40 series, visible 8GB)
  • Memory: ≥16GB (recommended over 32GB to load large models)
  • Storage: at least 50 GB space reserved (integrated package + model file)

2. Software dependency

  • Operating systems: Windows 11 64 (2H2 or updated)
  • CUDA Drive: CUDA 12.1 (adopted)nvidia-smiCommand to check driver version)
  • Counterpression ToolInstallation of 7-Zip or WinRAR to depressurize.7zDocumentation

Integration package deployment

1. Downloads and depressures

  1. Download integration package:
    ComfyUI Windows Portal (NVIDIA GPU)
  2. Unpressure to unsystematic disks (e. g.D:\ComfyUI_windowsThis post is part of our special coverage Syria Protests 2011.Ensure path does not contain Chinese or spacesI don't know.

Download models and CLIP files

  1. Main Model Download:
  1. Access stable-diffusion-3.5-large model page
  2. Downloadstable-diffusion-3.5-large.safetensors
  4. Storage path:ComfyUI_windows\ComfyUI\models\checkpoints
  5. CLIP text encoder download(https://huggingface.co/stabilityai/stable-diffusion-3.5-large/tree/main/text encoders)
    • clip_g.safetensors
    • clip_l.safetensors
    • t5xxl_fp16.safetensors
    • Access CLIP File Directory
    • Download the following three documents:
    • Storage path:ComfyUI_windows\ComfyUI\models\clip

3. Share Model Path (optional)

To share models with other Stable Diffusion tools (e.g. WebUI):

  1. OpenComfyUI_windows\ComfyUI\extra_model_paths.yaml.exampledelete.exampleThe suffix.
  2. Modify the contents of the document to the shared model directory:base_path: "C:\path\to\your\shared_models"  # 替换为实际路径
    checkpoints: models/checkpoints
    clip: models/clip

III. Launch and configuration

1. Run integration packages

  • Double clickComfyUI_windowsIn Directoryrun_nvidia_gpu.bat
  • ​
  • Initial start will automatically install dependency, and browser access after completionhttp://127.0.0.1:8188

2. Loading official sample workflows

  1. Download official workflow template: SD3.5L example workflow.json (https://huggingface.co/statilityai/table-diffiusion-3.5-large/resolve/main/SD3.5L example workflow.json?download=true)
  2. In the ComfyUI interface:
    • Click on the top right cornerLoadSelect a downloaded JSON file
    • Automatically load pre-defined nodes (including CHeckpoint, CLIP, samplers, etc.)

IV. BASIC USE PROCESS

1. Generation of images

  1. Checking work streamsCheckpointLoaderNode Selectedstable-diffusion-3.5-largeModels.
  2. Other Organiser
    • Inverted: input description (e.g. "A cyberpunk cityscape, nonlights, Rain, 8k")
    • Inverted hint: input should exclude content (e.g. "text, watermark, blurry")
  3. ClickQueue PromptStart generation, save results toComfyUI_windows\ComfyUI\output

2. Proposed parameters adjustments

ParametersRecommended valueAnnotations
Resolution1024x1024Fit SD3.5-Large training specifications
Sample steps20-30Recommended DPM++2M SDE or Euler
CFG Scale5-9The higher the value, the closer the hint.
Random Feeds-1 (random)Fixed seed recoverable result

Common issues

1. Model load failure

  • SymptomOther OrganiserCLIP model not foundorInvalid checkpoint
  • Solve:
    1. Check if CLIP files are fully storedmodels/clipContents.
    2. Confirm that the main model file is not redundant suffixed (e.g. duplicated).safetensorsI don't know.

2. Visibility shortfalls

  • Symptom: Crash or miss at generationCUDA out of memory
  • Solve:
    1. Modifyrun_nvidia_gpu.bat, add--lowvramparameter.
    2. Reduce resolution to 768 x 768 or enable TAESD light Quantification VAE.

VI. EXTENSION OF RESOURCES

  • Plugin Installation: adoptedComfyUI-ManagerInstall a ControlNet preprocessor or image magnification node.
  • Workstream communityOther Organiser

Download links to "CommyUI windows portable nvidia.7z"
Link: https://pan.quark.cn/s/f514c44a027