| name | node-reference |
| description | Common ComfyUI node types and workflow patterns. Use when suggesting nodes or explaining workflows. |
Common Node Types Reference
This reference helps you suggest appropriate nodes for user needs.
Loaders & Models
- CheckpointLoaderSimple: Loads SD model (outputs: MODEL, CLIP, VAE)
- LoraLoader: Adds LoRA to model (inputs: MODEL, CLIP)
- VAELoader: Loads separate VAE
- ControlNetLoader: Loads ControlNet models
Conditioning (Prompts)
- CLIPTextEncode: Positive/negative prompts (input: CLIP, text)
- ConditioningCombine: Combines multiple conditionings
- ConditioningSetArea: Area conditioning for inpainting
Latents
- EmptyLatentImage: Creates blank latent (specify width/height)
- VAEEncode: Image to latent (input: VAE, IMAGE)
- VAEDecode: Latent to image (input: VAE, LATENT)
- LatentUpscale: Upscale latents
Sampling
- KSampler: Standard sampler (inputs: MODEL, LATENT, positive, negative)
- KSamplerAdvanced: More control over sampling
- SamplerCustom: Custom sampling pipeline
Image Processing
- LoadImage: Load from disk
- SaveImage: Save to disk
- ImageScale: Resize images
- ImageUpscaleWithModel: AI upscaling
- ImageBatch: Combine images into batch
Utilities
- PrimitiveNode: Store constants/values
- Reroute: Clean up connections
- Note: Add notes/comments
Common Workflows
Important: This section is a conceptual reference only. Do not apply or generate workflows from these patterns unless you have followed the workflow-guardrails skill and verified installed nodes and models.
Basic txt2img:
CheckpointLoaderSimple → KSampler → VAEDecode → SaveImage
↓
CLIPTextEncode (positive)
CLIPTextEncode (negative)
EmptyLatentImage
img2img:
CheckpointLoaderSimple → KSampler → VAEDecode → SaveImage
LoadImage → VAEEncode ↗
Common Widget Names
Use these with setNodeWidgetValue. Call getWorkflowInfo with includeNodeDetails: true to confirm exact names.
- KSampler: seed, steps (number), cfg (number), sampler_name (string: "euler", "euler_ancestral", "dpmpp_2m", etc.), scheduler (string: "normal", "karras", "exponential", etc.), denoise (number: 0.0–1.0)
- EmptyLatentImage: width (number), height (number), batch_size (number)
- CLIPTextEncode: text (string) — use
fillPromptNode for convenience
- SaveImage: filename_prefix (string)
- KSamplerAdvanced: add_noise ("enable"/"disable"), noise_seed, steps, cfg, sampler_name, scheduler, start_at_step, end_at_step, return_with_leftover_noise ("disable"/"enable")
- ImageScale: upscale_method (string), width (number), height (number), crop (string)
Prompt Writing Guide
When filling CLIPTextEncode text widgets:
Positive prompt structure: subject, style/medium, quality modifiers, lighting/mood
- Example: "a fluffy orange cat sitting on a windowsill, digital painting, highly detailed, warm afternoon sunlight"
Negative prompt terms: things to avoid in the generation
- Example: "blurry, low quality, deformed, ugly, watermark, text, oversaturated"
Tips:
- Be specific and descriptive for positive prompts
- Separate concepts with commas
- Quality terms like "masterpiece, best quality, highly detailed" can help
- For negative, list common artifacts to avoid