<pre><code>Hey! 👋 This is the realism anchor of the Bonkaiii Spectrum. 

### ⚙️ Recommended settings
| Setting | Value |
|---------|-------|
| **CFG** | **5.0** (range 3.0–7) |
| **Sampler** | Euler a |
| **Steps** | 30-40 |
| **Resolution** | 832×1216 portrait, 1216×832 landscape |
| **VAE** | baked in ✓ |
| **Hires** | optional, 1.5× with 4x_NMKD-Siax_200k or similar works great |

⚠️ **CFG warning:** keep your prompt clean (no booru-style quality stack) at 5+. If you want to use heavy quality tags like `masterpiece, best quality, very aesthetic`, drop CFG to 3.5-4 to avoid burning.

### ⚠️ IMPORTANT — skip the booru quality stack
This is a **photoreal merge**, not an anime checkpoint. If you pile on `masterpiece, best quality, very aesthetic, absurdres, hyperrealistic` and parenthesized weight boosts on every quality tag, it WILL over-condition and burn your output (hyper-saturated, posterized, neon nightmare). Trust the merge — minimal quality conditioning or none at all is the way.

### 🎯 Suggested negative prompt
```
lowres, bad anatomy, bad hands, missing fingers, extra digits, deformed, blurry, jpeg artifacts, watermark, signature, text
```

Just the anatomy floor. No more.

### 💡 Tip
Pair this with ADetailer for full-body shots — the SDXL architecture struggles to render fine facial detail at 80-100px face heights in 832×1216 full-body framing. ADetailer fixes that beautifully.</code></pre><p></p>

Bonkaiii_realism

<p>v2 realism is way more stable than v1 was. The neon nightmare it could hit at CFG 5+? Pretty much gone. Two LoRAs at low weights, doing the work of three things at once.</p>

Files with SHA256: aca2d402e59b4b832db72a806bd7a6a876162a7cc4a320154f88d8c2fefddc55