This lora adds more realistic (not always...) elements to generative chinese food contents. It is trained with 20 royalty-free images with tag style labels. The food ingredients include pork, duck, shrimps, crab, seafood, tofu, vegetable...

Example software: webui forge

Model: original dev, schnell and their nf4 versions.

Diffusion in Low Bits: Automatic (fp16 LoRA)

Prompt: a dish of {prompt}, chinese cuisine, food advertisement photography

Lora strength: 0.5-1.

Sampler: Euler simple/beta

Steps: 20-30 for dev model, 4 for schnell model.

Training parameter:

{
  "engine": "kohya",
  "unetLR": 0.0005,
  "clipSkip": 1,
  "loraType": "lora",
  "keepTokens": 0,
  "networkDim": 2,
  "numRepeats": 20,
  "resolution": 768,
  "lrScheduler": "cosine_with_restarts",
  "minSnrGamma": 5,
  "noiseOffset": 0.1,
  "targetSteps": 1000,
  "enableBucket": true,
  "networkAlpha": 16,
  "optimizerType": "AdamW8Bit",
  "textEncoderLR": 0,
  "maxTrainEpochs": 10,
  "shuffleCaption": false,
  "trainBatchSize": 4,
  "flipAugmentation": false,
  "lrSchedulerNumCycles": 3
}