Purpose: Negative / Inverted colors are a bit of a rare art form but its one I always had a soft spot for. This LoRA hopes to capture a bit of that. Sure you can generate an image normally and run it through a color inverter afterwards but where's the fun in that? Mess with the weightings and combine with other LoRAs to create something unique. Recommended Weights: 0.0 - 0.6: You might see a little bit of negative color here an there. Not much but might add some flavor. 0.6 - 085: This is the "fun" zone IMO. The image becomes almost pseudo-negative. Some colors are negative some aren't. Produces interesting results. 0.85 - 1: Colors will mostly become true negative. You might still need to prompt certain things like the sky color etc. Extra Notes: This is my first time trying something like this. I just sort of wanted to see if it was possible to train a LoRA to recognize a specific color pattern. A little surprised it sort of worked. I used a lot of different weightings in my sample pics, still not sure why the exact weights don't show on Comfyui. Its trained entirely on AI generated images that have been run through a negative filter. I found pretty quickly that AI images tend to look a lot better after the filter was applied. I'm sure there's a good reason for that but I can't narrow it down to one. Images can be a little blurry sometimes depending on your checkpoint. Not sure if I overtained it or if some other reason is causing that.

AnthroColor - Negative / Inverted Colors

Trained on 829 images: { "engine": "kohya", "unetLR": 1, "clipSkip": 2, "loraType": "lora", "keepTokens": 1, "networkDim": 16, "numRepeats": 2, "resolution": 1024, "lrScheduler": "cosine", "minSnrGamma": 5, "noiseOffset": 0.1, "targetSteps": 8290, "enableBucket": true, "networkAlpha": 8, "optimizerType": "Prodigy", "textEncoderLR": 1, "maxTrainEpochs": 20, "shuffleCaption": true, "trainBatchSize": 4, "flipAugmentation": true, "lrSchedulerNumCycles": 3}

Files with SHA256: 4943af40b2f400b59478fb0133160a28d6a134da32cd67f1b2e5adce1cdcaf3c