this was an experiment to see if i could guide the position of characters with a caption. it doesn't really work lol. also makes everything look ugly. u probably shouldn't use this on civitai or u'll lose a lot of buzz for nothing lmaobut if u gen dozens and dozens of times sometimes it at least gets the positions right. u can probably fix the artifacts by doing img2img without the lora or with the lora at a weaker strength or inpainting. all of these things are easier to do locally<h2 id="v1">v1</h2>trigger: boy_character caught crossdressing by girl_character, sandwiched between left_character and right_character, bra, skirt, indoors, holding panties, clenching hand, 1girl, 3boys

Caught Crossdressing (Danshi Koukousei no Nichijou)

trained on 8 images<pre><code>{
 "engine": "kohya",
 "unetLR": 0.0001,
 "clipSkip": 1,
 "loraType": "lora",
 "keepTokens": 2,
 "networkDim": 32,
 "numRepeats": 15,
 "resolution": 1024,
 "lrScheduler": "cosine_with_restarts",
 "minSnrGamma": 5,
 "noiseOffset": 0,
 "targetSteps": 1200,
 "enableBucket": true,
 "networkAlpha": 32,
 "optimizerType": "Adafactor",
 "textEncoderLR": 0.00005,
 "maxTrainEpochs": 40,
 "shuffleCaption": true,
 "trainBatchSize": 4,
 "flipAugmentation": false,
 "lrSchedulerNumCycles": 4
}</code></pre>

Files with SHA256: 14517785d056ded32a2b328349dbda37fc6c2b67957503ffaff3739076072ca1