Surrealistic portraits in style of Naoto Hattori (LoRA for SD1.5)

Description

You can get this embedding for $0 here. Also you can download and test it for FREE via CivitAI ; or via TensorArt.

LoRA for SD1.5 trained with CivitAI on 120 (60 originals + 60 mirrored) 680×680 surrealistic mostly female (sometimes doll, android, unisex) portraits in style of Naoto Hattori with manually fixed BLIP2 captions.

Training values (all default except resolution): { “unetLR”: 0.0005, “clipSkip”: 1, “loraType”: “lora”, “keepTokens”: 0, “networkDim”: 32, “numRepeats”: 2, “resolution”: 680, “lrScheduler”: “cosine_with_restarts”, “minSnrGamma”: 5, “targetSteps”: 400, “enableBucket”: true, “networkAlpha”: 16, “optimizerArgs”: “weight_decay=0.1”, “optimizerType”: “AdamW8Bit”, “textEncoderLR”: 0.00005, “maxTrainEpochs”: 10, “shuffleCaption”: false, “trainBatchSize”: 6, “flipAugmentation”: false, “lrSchedulerNumCycles”: 3 }

LoRA should understand:

1. Several tips of backgrounds/environments: “black background, simple background, isolated on black” OR “black background, starry sky” OR “surrealistic landscape, grassy area on the foreground and mountains on the background” OR “sky”.
2. Difference between open and closed eyes: “open eyes, big eyes with elegant eyelashes” OR “closed eyes with long eyelashes, sleeping, in a dream”.
3. Difference between cuteness and doom: “pretty face” (for cute portraits) OR “dead woman” (for horrifying portraits).
4. Special tags phrases: “liquid flash” (for flesh or hair acting like a liquid, unfortunately mistake in word ‘flAsh’ is already trained into model), “head flying in the sky” (for head without a body, usually for sky and landscape background), “head flying in the void” (for head without a body, usually for black background), “skin like a paper, skin like a bark, skin turns into ribbons and paper cuts” (for most common futures of Naoto Hattori style)