CFG 2.8 vs 3.0 vs 4.2 Reference Behavior

CFG 2.8 (Soft/Cohesive)	CFG 3.0 (Balanced)	CFG 4.2 (Sharp/Stylized)
Prompt: masterpiece, ultra quality, juri han, high contrast lighting, stylized anatomy emphasis, sharp focus Behavior Notes Softer anatomy transitions More natural lighting rolloff Reduced edge sharpening CFG: 2.8	Prompt: masterpiece, ultra quality, jinx (league of legends), neon graffiti alley, dynamic pose, cinematic lighting Behavior Notes Best prompt fidelity balance Stable composition control CFG: 3.0	Prompt: masterpiece, ultra quality, juri han, high contrast lighting, stylized anatomy emphasis, sharp focus Behavior Notes Increased stylization pressure Stronger edge definition Higher risk of artifacting CFG:4.2

Example Outputs

🧪 Model Variant Comparison — CFG 3.2 (A1 vs D3 vs D4)

A1 — CFG 3.2 (Balanced / Softer Stylization)
A1 • CFG 3.2	A1 • CFG 3.2	A1 • CFG 3.2
D3 — CFG 3.2 (Refined Detail / Balanced Output)
D3 • CFG 3.2	D3 • CFG 3.2	D3 • CFG 3.2
D4 — CFG 3.2 (High Detail / Strong Stylization)
D4 • CFG 3.2	D4 • CFG 3.2	D4 • CFG 3.2

Prompting Guidance

Workflows

• workflows/gz_ultimate_v3.json

Repository Structure

• examples/ → curated outputs
• workflows/ → ComfyUI workflows
• release_candidates/ → staged checkpoints

Release System

• Each RC reflects controlled experimentation
• Iteration focuses on output stability and consistency
• Final selection is based on visual coherence across seeds

License

• No redistribution of weights
• No derivative model merges
• No illegal or harmful use https://huggingface.co/spaces/CompVis/stable-diffusion-license

Links

• Hugging Face: https://huggingface.co/graizelle/graizelle_lustmix_v2.0
• CivitAI: https://civitai.com/models/1995001/graizelles-lustmix