[ Work in Progress ]Updated to Source 2: Better image quality, revised tagging, character and location tags<ul><li>The trigger words are: Source Engine, 3D,</li></ul>Characters that are baked in:<ul><li>Alyx is triggered by: AlyxV, Dark Skin, Jacket</li><li>Judith can be triggered by: JudithM, sweater, hair ornament</li><li>Female 01 can be triggered by: Female 01, hair bun</li><li>Female 02 can be triggered by: Female 02, ponytail</li><li>Combine are triggered with: Combine, Combine Mask, Combine uniform</li></ul>Tags that assist with location:<ul><li>indoors</li><li>red couch</li><li>rug</li><li>window</li><li>laboratory</li><li>facility</li><li>outdoors</li><li>gm_construct</li><li>blue sky</li><li>clouds</li><li>grass</li><li>building</li><li>scenery</li><li>fence</li><li>courtyard</li><li>concrete wall</li></ul>If you're sometimes getting bigger heads, put this in negative:<ul><li>source_anime</li></ul>To get results akin to the example images I recommend using Latent Upscaling with nearest-exact at 1.50.I plan on updating this frequently but I've never made LoRAs before and that costs buzz lolThanks for downloading, excited to see what y'all post!

Source Engine - Style LoRA

First ever LoRA trained on Civit, I absolutely plan to rework and perfect this. At this point im just checking settings within the model editor for further upgrades. This was trained on 50 Epochs, this is Epoch 37. Keeping it at 0.9 strength seemed to work for me. I do know that the generations get scuffed but using Latent Upscaling I get pretty damn accurate pictures to the game.

Files with SHA256: 2273a38ba8bab2e049cd2d0b8cff804e87a58bc38cd1fc8d842f4a6bc4a4e344