Phase 6: ComfyUI — Workflow Engine · 60 min · Python · ComfyUI

Workflow Design Patterns

Base+refiner, LoRA stacking, ControlNet, upscaling, batch, image-to-video — these patterns are the building blocks of every ComfyUI production workflow.

Hiring signal: Workflow design pattern knowledge (base+refiner, LoRA stacking, ControlNet, upscaling) is tested in ComfyUI workflow challenges — you need to design and build these from scratch.

What you will learn

Build a base generation workflow: checkpoint → prompt → KSampler → VAE decode → save
Build a refiner pipeline: base model generates latent → refiner model improves quality
Build a LoRA stacking workflow: multiple LoRA loaders with weight blending
Build an upscaling workflow: generate → latent upscale → high-res KSampler → ESRGAN

The Problem

A team knows ComfyUI basics — load checkpoint, encode prompt, sample, save. But they need:

Higher quality: base model + refiner model for detail enhancement
Style control: multiple LoRAs blended together
Structural control: ControlNet to guide composition from a reference image
Higher resolution: upscaling without losing quality
Batch generation: multiple images from one workflow

They don't know the design patterns that solve these problems. Workflow design pattern knowledge is tested in ComfyUI workflow challenges — you need to design and build these from scratch.

What you'll build
Build a base+refiner pipeline, LoRA stacking workflow, ControlNet workflow, and upscaling workflow. Understand batch generation patterns. These are the building blocks of every production ComfyUI workflow.

Pattern 1: Base + Refiner

The base model generates a latent at low resolution, the refiner model improves detail:

LoadCheckpoint (base) → CLIPTextEncode → KSampler (base, denoise=1.0)
    → KSampler (refiner, denoise=0.3) → VAEDecode → SaveImage
LoadCheckpoint (refiner) → CLIPTextEncode (refiner)

def build_base_refiner_workflow(prompt: str, negative: str = "",
                                base_ckpt: str = "sd_xl_base_1.0.safetensors",
                                refiner_ckpt: str = "sd_xl_refiner_1.0.safetensors",
                                seed: int = 42) -> dict:
    """Build a base+refiner workflow."""
    return {
        # Base checkpoint
        "1": {"class_type": "CheckpointLoaderSimple",
              "inputs": {"ckpt_name": base_ckpt}},
        # Refiner checkpoint
        "2": {"class_type": "CheckpointLoaderSimple",
              "inputs": {"ckpt_name": refiner_ckpt}},
        # Positive prompt (base)
        "3": {"class_type": "CLIPTextEncode",
              "inputs": {"clip": ["1", 1], "text": prompt}},
        # Negative prompt (base)
        "4": {"class_type": "CLIPTextEncode",
              "inputs": {"clip": ["1", 1], "text": negative}},
        # Positive prompt (refiner)
        "5": {"class_type": "CLIPTextEncode",
              "inputs": {"clip": ["2", 1], "text": prompt}},
        # Negative prompt (refiner)
        "6": {"class_type": "CLIPTextEncode",
              "inputs": {"clip": ["2", 1], "text": negative}},
        # Empty latent
        "7": {"class_type": "EmptyLatentImage",
              "inputs": {"width": 1024, "height": 1024, "batch_size": 1}},
        # Base KSampler (full denoise)
        "8": {"class_type": "KSampler",
              "inputs": {"seed": seed, "steps": 25, "cfg": 8.0,
                         "sampler_name": "dpmpp_2m", "scheduler": "karras",
                         "denoise": 1.0, "model": ["1", 0],
                         "positive": ["3", 0], "negative": ["4", 0],
                         "latent_image": ["7", 0]}},
        # Refiner KSampler (low denoise — only refine detail)
        "9": {"class_type": "KSampler",
              "inputs": {"seed": seed, "steps": 25, "cfg": 8.0,
                         "sampler_name": "dpmpp_2m", "scheduler": "karras",
                         "denoise": 0.3, "model": ["2", 0],
                         "positive": ["5", 0], "negative": ["6", 0],
                         "latent_image": ["8", 0]}},
        # VAE Decode
        "10": {"class_type": "VAEDecode",
               "inputs": {"samples": ["9", 0], "vae": ["1", 2]}},
        # Save
        "11": {"class_type": "SaveImage",
               "inputs": {"images": ["10", 0], "filename_prefix": "refined"}},
    }

Key: The refiner's denoise=0.3 means it only changes 30% of the image — refining detail without altering composition.

Unlock the full lesson

You've read the first 2 sections. The rest of this lesson covers Pattern 2: LoRA Stacking, Pattern 3: ControlNet, Pattern 4: Upscaling, Pattern 5: Batch Generation, Key Takeaways, What's Next — plus a hands-on lab, quiz, and project artifact.

Create a free account to unlock Phase 0 and Phase 1 of every course — no credit card.

Key terms in this lesson

LoRA (Low-Rank Adaptation)