ComfyUI Workflow Library: Complete Guide & Expert Strategies

1. Loader Nodes

Load checkpoints, LoRAs, VAEs, and other models into memory.

• • CheckpointLoaderSimple: Loads base SD model (SDXL, SD 1.5, etc.)
• • LoraLoader: Adds style/character LoRAs on top of checkpoint
• • VAELoader: Loads VAE for encoding/decoding latents to pixels

2. Conditioning Nodes

Convert text prompts into embeddings the model understands.

• • CLIPTextEncode: Processes positive/negative prompts into conditioning vectors
• • ConditioningCombine: Merges multiple conditioning inputs
• • ConditioningSetArea: Applies conditioning to specific image regions

3. Sampling Nodes

The core denoising process that generates images.

• • KSampler: Basic sampler with seed, steps, CFG, sampler method selection
• • KSamplerAdvanced: Add noise levels, denoise strength control
• • Common samplers: DPM++ 2M Karras, Euler a, DDIM, UniPC

4. Latent Nodes

Create and manipulate the latent space (compressed image representation).

• • EmptyLatentImage: Creates blank latent at specified resolution
• • VAEEncode: Converts pixel image to latent space
• • LatentUpscale: Scales latent before decoding for higher resolution

5. Image Processing Nodes

Post-process final outputs or prepare inputs.

• • VAEDecode: Converts latent back to viewable pixel image
• • ImageScale: Resize images with various algorithms (lanczos, nearest, etc.)
• • SaveImage: Exports final output to disk with metadata

Connection Types:

• • MODEL: Purple - Checkpoint data flows from loaders to samplers
• • CONDITIONING: Yellow - Text embeddings flow to sampler positive/negative inputs
• • LATENT: Pink - Compressed images flow between latent operations
• • IMAGE: Green - Pixel data flows through image processing nodes
• • VAE: Red - VAE model flows to encode/decode nodes

Execution Order:

ComfyUI automatically determines execution order by analyzing dependencies. Nodes execute only when all required inputs are available. This allows parallel execution of independent branches.