This is a simple kohya extract (256 dim and 64 dim for the Lite version) of the 1.5 model trained with offset noise by <a target="_blank" rel="ugc" href="https://www.crosslabs.org//blog/diffusion-with-offset-noise">Cross Lab </a>and should work well with most 1.5 based models.Applying this lora will help you generate more realistic exposure for your generations, especially in pretty contrasty scenarios, where the original 1.5 would have tried to create a boring even exposure or overexposed/over saturated results to compensate the dark parts. Note that it will probably change composition<h2>Instructions :</h2><ul><li>Placing it at the start of the prompt seems to help</li><li>It's fairly strong, you can dial it back if need be</li><li>I recommend using dark themed prompts of course</li></ul>Here's an example with <a target="_blank" rel="ugc" href="https://civarchive.com/models/4201/realistic-vision-v20">Realistic Diffusion 2.0</a><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/f1b79aff-0d72-4af7-b23f-7a68cd24f890/width=1024/00010-1232856568.0-dreamy%20close-up%20underexposed%20(portrait%20photography_1.3)%20of%20a%20young%20blonde%20woman%20in%20a%20field%20(at%20night_1.2),%20silhouette,%20(dark,_1.jpeg" alt="00010-1232856568.0-dreamy close-up underexposed (portrait photography_1.3) of a young blonde woman in a field (at night_1.2), silhouette, (dark,_1.png" />

1.5 Offset Noise Extract

Just a lighter Lora, probably less precise but doesn't even seem so !

Files with SHA256: 8491f79879403562a4ba3d0f62c02fde988888a24f467f7a14796dd833dd6bcd