CFG Scale Optimal Settings: Complete Guide 2025
Master CFG scale settings for Stable Diffusion, SDXL, Flux, and SD 3.5. Learn how classifier-free guidance affects quality and find optimal values for each model.
CFG Scale (Classifier-Free Guidance) controls how closely AI follows your prompt versus generating freely. Too low and the model ignores your prompt. Too high and images become oversaturated and artificial.
Quick Answer: Optimal CFG varies by model: SD 1.5 uses 7-11, SDXL works best at 5-8, SD 3.5 prefers 3.5-5, and Flux uses 25-50 (different scale). Start in these ranges and adjust based on results.
- SD 1.5: 7-11 (default 7)
- SDXL: 5-8 (default 7)
- SD 3.5: 3.5-5 (lower than SDXL!)
- Flux: 25-50 (different scale entirely)
- Pony/Illustrious: 5-7
What Is CFG Scale?
CFG stands for Classifier-Free Guidance. It determines the balance between:
- Unconditional generation: What the model would create without your prompt
- Conditional generation: What your prompt specifically requests
Low CFG (1-4): More creative freedom, may ignore prompt details Medium CFG (5-9): Balanced prompt adherence and creativity High CFG (10+): Strict prompt following, may cause artifacts
How CFG Affects Image Quality
Low CFG (1-4)
- Softer, more natural images
- Colors are muted
- Prompt may be partially ignored
- Good for: Abstract art, soft portraits, dream-like effects
Medium CFG (5-9)
- Balanced quality
- Good prompt adherence
- Natural color saturation
- Good for: Most use cases, general generation
High CFG (10-15)
- Strong prompt adherence
- Increased saturation
- More contrast
- Risk of artifacts
- Good for: Precise concepts, specific details
Very High CFG (15+)
- Often causes artifacts
- Oversaturated colors
- Unnatural appearance
- Generally avoid unless specifically needed
Model-Specific Recommendations
Stable Diffusion 1.5
Optimal Range: 7-11 Default: 7
SD 1.5 is forgiving with CFG. The classic "7" works well for most cases.
Photorealistic: 7-8
Artistic/Stylized: 8-10
Anime: 7-9
Abstract: 4-6
SDXL
Optimal Range: 5-8 Default: 7
SDXL responds more strongly to CFG. Lower values often produce better results.
Photorealistic: 5-7
Artistic: 6-8
Detailed scenes: 7-8
Portraits: 5-6
SD 3.5
Optimal Range: 3.5-5 Default: 4.5
SD 3.5 uses significantly lower CFG than previous versions. This is intentional—the architecture responds differently.
General use: 4-5
Photorealistic: 3.5-4.5
Artistic: 4.5-5.5
Text rendering: 4-5
Flux
Optimal Range: 25-50 Default: ~30
Flux uses a completely different guidance scale. Don't apply SD values!
Standard generation: 30-35
Strong prompt following: 40-50
Creative/loose: 20-25
Pony Diffusion / Illustrious
Optimal Range: 5-7 Default: 6
Anime-focused models often prefer moderate CFG.
Free ComfyUI Workflows
Find free, open-source ComfyUI workflows for techniques in this article. Open source is strong.
Standard anime: 6-7
Soft style: 5-6
Detailed: 7-8
CFG Scheduling
Advanced technique: vary CFG during generation.
High-to-Low Schedule:
- Start with high CFG (establish composition)
- Reduce CFG later (refine details naturally)
In ComfyUI: Use CFG scheduling nodes to implement this. Example:
- Steps 0-10: CFG 9
- Steps 10-20: CFG 7
- Steps 20-30: CFG 5
Troubleshooting CFG Issues
Issue: Images look washed out
Cause: CFG too low Solution: Increase CFG by 1-2 points
Issue: Oversaturated, artificial colors
Cause: CFG too high Solution: Lower CFG by 2-3 points
Issue: Prompt being ignored
Cause: CFG too low Solution: Increase CFG, also check prompt quality
Want to skip the complexity? Apatero gives you professional AI results instantly with no technical setup required.
Issue: Strange artifacts/distortions
Cause: CFG too high or conflicting with model Solution: Lower CFG, try different sampler
Issue: Image lacks contrast
Cause: CFG may be in awkward range Solution: Try both higher and lower to compare
CFG and Other Parameters
CFG interacts with other settings:
Steps:
- Lower steps may need slightly higher CFG
- More steps can use lower CFG
Sampler:
- Euler: Standard CFG ranges work
- DPM++: May need slightly lower CFG
- Ancestral samplers: More sensitive to CFG
Prompt Length:
Join 115 other course members
Create Your First Mega-Realistic AI Influencer in 51 Lessons
Create ultra-realistic AI influencers with lifelike skin details, professional selfies, and complex scenes. Get two complete courses in one bundle. ComfyUI Foundation to master the tech, and Fanvue Creator Academy to learn how to market yourself as an AI creator.
- Longer prompts: May need slightly lower CFG
- Short prompts: May need higher CFG for adherence
Practical Testing Approach
Find your optimal CFG:
- Generate baseline at default CFG (7 for SD/SDXL)
- Test low (CFG 4-5) — check if still follows prompt
- Test high (CFG 10-12) — check for artifacts
- Narrow range based on results
- Fine-tune in 0.5 increments
Most workflows settle on a CFG sweet spot for their specific model/style combination.
CFG for Different Content Types
Portraits: Lower CFG (5-7) for natural skin tones
Landscapes: Medium CFG (7-9) for good detail
Abstract Art: Lower CFG (3-5) for creative interpretation
Product Photography: Higher CFG (8-10) for accurate details
Anime/Illustration: Medium CFG (6-8) for style consistency
Frequently Asked Questions
Why does Flux use such high CFG numbers?
Flux uses a different guidance implementation. The scale isn't comparable to SD/SDXL—it's not "more intense," just different math.
Can CFG be 0?
Yes, but the model will completely ignore your prompt, generating random content.
Does CFG affect generation speed?
No, CFG doesn't impact computation time directly.
Should I change CFG when using LoRAs?
Sometimes. Strong LoRAs may need slightly lower CFG to prevent over-emphasis.
What's the relationship between CFG and negative prompts?
CFG amplifies both positive and negative prompts. Higher CFG = stronger negative prompt effect too.
Conclusion
CFG scale is one of the most impactful parameters for image quality. The key insight is that optimal CFG varies significantly between models—SD 3.5's ideal 4.0 would be too low for SD 1.5.
Start with recommended defaults for your model, then adjust based on results. When in doubt, try both higher and lower values—the difference is often dramatic and helps you find the sweet spot for your specific use case.
Ready to Create Your AI Influencer?
Join 115 students mastering ComfyUI and AI influencer marketing in our complete 51-lesson course.
Related Articles
10 Best AI Influencer Generator Tools Compared (2025)
Comprehensive comparison of the top AI influencer generator tools in 2025. Features, pricing, quality, and best use cases for each platform reviewed.
AI Adventure Book Generation with Real-Time Images
Generate interactive adventure books with real-time AI image creation. Complete workflow for dynamic storytelling with consistent visual generation.
AI Background Replacement: Professional Guide 2025
Master AI background replacement for professional results. Learn rembg, BiRefNet, and ComfyUI workflows for seamless background removal and replacement.