<p><strong>Clayflash</strong> is a realism-focused style LoRA designed to push images toward a direct camera-flash, candid snapshot look. It adds harder frontal lighting, flash falloff, sharper subject separation, darker backgrounds, mild blown-highlight behavior, and that slightly raw “caught by the camera” feel you get from disposable-camera, party-photo, nightlife, and casual flash photography.</p><p>This LoRA was trained with a CyberRealistic/Pony realism influence, so it is intentionally not a sterile flash preset. It tends to bring along a bit of CyberRealistic-style skin, face rendering, contrast, and composition flavor while still staying flexible enough for normal prompts.</p><h2 id="trigger-word">Trigger Word</h2><p>Use:</p><p><code>clayflash</code></p><h2 id="suggested-strength">Suggested Strength</h2><p>Start around:</p><p><code>0.3 – 0.7</code></p><p>Higher strengths can make the flash effect stronger, but may also push the image too far and it just falls apart</p><h2 id="recommended-prompting">Recommended Prompting</h2><p>Example:</p><p><code>score_9, score_8_up, score_7_up, clayflash, realistic photo of a woman standing in a kitchen, casual outfit, looking at camera, candid snapshot, indoor background</code></p><p>Other useful prompt terms:</p><p><code>candid snapshot, direct flash, hard shadows, nightlife photo, party photo, flash photo, realistic skin, looking at camera, indoor background</code></p><h2 id="negative-prompt">Negative Prompt</h2><p><code>score_4, score_5, score_6, blurry, low quality, bad anatomy, bad hands, deformed hands, extra fingers, text, watermark, logo, distorted face, plastic skin</code></p><h2 id="notes">Notes</h2><p>This LoRA works best when used for realistic portraits, candid indoor photos, party/nightlife shots, casual fashion photos, mirror/selfie-style images, and flash-lit object or vehicle photos. It may produce stronger results on Pony/realism-based checkpoints, especially CyberRealistic-style models.</p><p>For best results, test different strengths. The sweet spot is usually lower than expected; around <code>0.4–0.6</code> often gives the flash-photo look without overpowering the base model.</p>

Files with SHA256: 5747ce3d9e4cef030245a0e7e58c799916e8e1e6b843984dad42dcf2861bb91c