ComfyUI LoRA Models: 13 Downloads for Realistic & NSFW Generation

Quick Reference

Detail Enhancers

Detail Tweaker (SD 1.5)

557,000+ downloads - the standard detail LoRA.

• Architecture: SD 1.5 | File size: 36 MB | VRAM: 4 GB+
• Trigger words: None - controlled by LoRA weight (-2 to +2)
• Download from CivitAI

Controls detail density through weight alone. Negative weights smooth, positive weights sharpen. If you're on SD 1.5, install this first.

Detail Tweaker XL

SDXL version - 384,000+ downloads.

• Architecture: SDXL | File size: 218 MB | VRAM: 6 GB+
• Trigger words: None - weight range -3 to +3
• Download from CivitAI

Same concept, wider weight range. Stack it with a style LoRA for the best results on SDXL checkpoints.

xl_more_art-full

2.9 million downloads - most downloaded SDXL LoRA on CivitAI.

• Architecture: SDXL | File size: 686 MB | VRAM: 8 GB+
• Trigger words: None required (use "Aerial" for aerial-style outputs)
• Recommended weight: 0.8-1.0
• Download from CivitAI

Boosts aesthetics and detail across the board. The 686 MB file is large enough to slow loading on older hardware, but the quality gain justifies it.

FLUX Image Upgrader

1.6 million downloads - best detail enhancer for Flux.

• Architecture: Flux (also has SDXL and SD 1.5 versions) | File size: 73 MB | VRAM: 12 GB+
• Trigger words: aidmaimageupgrader
• Recommended weight: 0.4 (range 0.15-1.0)
• Download from CivitAI

Fixes detail and contrast at low CFG values. Start at weight 0.4 - going higher can oversaturate.

Add More Details (SD 1.5)

Lightweight 9 MB alternative to Detail Tweaker.

• Architecture: SD 1.5 | File size: 9 MB | VRAM: 4 GB+
• Trigger words: None - weight 0.5 to 1.0
• Download from CivitAI

By Lykon. Try negative weights for softening effects. Good as a second detail LoRA alongside Detail Tweaker.

Speed LoRAs

LCM & TurboMix

1 million downloads - 8-step SDXL generation in just 12 MB.

• Architecture: SDXL | File size: 12 MB | VRAM: 6 GB+
• Trigger words: None
• Settings: Weight ~1.0, CFG 2.0, Steps 6-8, Euler a sampler
• Download from CivitAI

Cuts SDXL generation from 20-30 steps down to 6-8 with minimal quality loss. The speed gain is massive for fast iteration.

SDXL Lightning

Official Stability AI distilled LoRA - 1 to 8 step generation at 1024px.

• Architecture: SDXL | File size: 376 MB (2-step variant) | VRAM: 6 GB+
• Trigger words: None
• Settings: Euler sampler, sgm_uniform scheduler, CFG 0
• Variants: 1-step, 2-step, 4-step, 8-step
• Download from CivitAI

The 4-step variant hits the best quality/speed balance. No negative prompts needed - skip them entirely with this LoRA.

Style & Realism

F.2 Midjourney V6.1 meets FLUX

830,000+ downloads - most downloaded Flux LoRA overall.

• Architecture: Flux (also Illustrious/SDXL versions) | File size: 317 MB | VRAM: 12 GB+
• Trigger words: aidmamj6.1 (v0.2+)
• Recommended weight: 0.3-1.0
• Download from CivitAI

Brings Midjourney-style aesthetics to Flux outputs. Start at 0.5 and adjust - too high makes everything look overprocessed.

Hyperrealistic (Pony / Illustrious)

Bridges anime and photorealism on Illustrious and Pony base models.

• Architecture: Illustrious (Pony v3.5 also available) | File size: 435 MB | VRAM: 6-8 GB
• Trigger words: Hype4realistic
• Download from CivitAI

Best for portraits and close-ups where you want semi-realistic output from an anime-focused model.

Touch of Realism

Photographic depth and lighting for SDXL portraits.

• Architecture: SDXL | VRAM: 6-8 GB
• Trigger words: touch-of-realismV2
• Download from CivitAI

Clean photographic quality boost. Pairs well with NSFW LoRAs for realistic portrait work.

Skin & Texture

Realistic Skin Texture

Cinematic skin realism - works on SD 1.5, SDXL, Flux, Pony, and Illustrious.

• Architecture: Multi-platform | File size: 162 MB | VRAM: Varies by base model
• Trigger words: incredibly lifelike, cinematic, fine texture
• Download from CivitAI

Adds pores, texture, and subsurface scattering. One file works across all major architectures - no separate downloads needed.

PhotorealTouch

Micro-texture realism - pores, vellus hair, subsurface scattering.

• Architecture: SDXL / ZImage Turbo / Pony | VRAM: 6-8 GB
• Trigger words: None - controlled by weight
• Settings: Weight 0.5-0.8, Steps 7-10, CFG 2.0-3.0
• Download from CivitAI

Goes further than Realistic Skin Texture on micro-detail. The tradeoff: it needs specific CFG and step settings to look right.

NSFW

FLUX NSFW Unlock

Most popular Flux NSFW LoRA - 130,000+ downloads.

• Architecture: Flux | File size: 18 MB | VRAM: 12-16 GB
• Trigger words: aidmaNSFWunlock
• Recommended weight: 0.9
• Download from CivitAI

Required for NSFW generation on base Flux.1 Dev. Stack with Touch of Realism or F.2 Midjourney for styled NSFW output.

How to Install LoRA Models in ComfyUI

1. Download the .safetensors file from the links above
2. Place it in ComfyUI/models/loras/
3. Restart or refresh ComfyUI
4. Add a Load LoRA node to your workflow
5. Connect Model + CLIP outputs from your checkpoint → Load LoRA inputs
6. Adjust strength_model and strength_clip to control intensity

Stacking rule: Chain Load LoRA nodes in series. Keep total combined strength under 2.0 and stick to 3 LoRAs max. Or use LocalForge AI for a pre-configured Forge setup where LoRA loading works out of the box.

What to Do Next

• Set up ComfyUI from scratch: ComfyUI NSFW Setup Guide
• Find checkpoints to pair with these LoRAs: Best NSFW Models for ComfyUI
• Explore Flux-specific models: Best Flux NSFW Models on CivitAI