Text-to-Image Workflow

Basic text-to-image generation workflow.

Workflow Overview

[CheckpointLoader] → [CLIPTextEncode(positive)] → [KSampler] → [VAEDecode] → [SaveImage]
                    ↓ CLIP
                    [CLIPTextEncode(negative)]

Step-by-Step

Step 1: Load Checkpoint

Node: CheckpointLoader
Select: Your model (e.g., v1-5-pruned-emaonly.safetensors)

Step 2: Encode Positive Prompt

Node: CLIPTextEncode
Input: clip from CheckpointLoader

Text: "masterpiece, best quality, a beautiful landscape with mountains and sunset"

Step 3: Encode Negative Prompt

Node: CLIPTextEncode
Input: clip from CheckpointLoader

Text: "blurry, low quality, distorted, ugly"

Step 4: Create Latent Image

Node: EmptyLatentImage
Width: 512
Height: 512
Batch size: 1

Step 5: Sample

Node: KSampler
Model: from CheckpointLoader
Positive: from CLIPTextEncode (positive)
Negative: from CLIPTextEncode (negative)
Latent: from EmptyLatentImage

Steps: 20
CFG: 8
Sampler: euler

Step 6: Decode

Node: VAEDecode
Samples: from KSampler
VAE: from CheckpointLoader

Step 7: Save

Node: SaveImage
Images: from VAEDecode
Filename prefix: output

Parameters Reference

Parameter	Recommended	Description
Width	512-768	Image width
Height	512-768	Image height
Steps	20-30	Sampling steps
CFG	7-10	Prompt influence
Sampler	euler	Algorithm

Prompt Tips

Quality Tags

# Always add
masterpiece, best quality, ultra detailed

# Style
digital art, photorealistic, 8k

Negative Prompt

blurry, low quality, distorted, ugly
bad anatomy, extra fingers, mutation

Advanced Options

SDXL

Use SDXL checkpoint for higher quality:

Workflow: Same structure, SDXL checkpoint
Parameters: 1024x1024, 30-50 steps

Higher Resolution

Width: 768, Height: 768
Steps: 30-40
CFG: 8-10

Next Steps

• Image-to-Image - Transform image
• Inpainting - Fix areas
• ControlNet - Add control