Description
Training on CivitAI settings
image count:170
numRepeats: 5
trainBatchSize: 5
engine: kohya
unetLR: 0.0005
clipSkip: 1
loraType: lora
keepTokens: 1
networkDim: 2
resolution: 512
lrScheduler: cosine_with_restarts
minSnrGamma: 5
noiseOffset: 0.1
targetSteps: 850
enableBucket: true
networkAlpha: 16
optimizerType: AdamW8Bit
textEncoderLR: 0
maxTrainEpochs: 5
shuffleCaption: true
flipAugmentation: false
lrSchedulerNumCycles: 3
FAQ
Comments (2)
Thank you for making this LoRA!
I really like how the colors get more vibrant with your CATalyst in comparison to standard Flux. As for the prompts... yeah i don't know either. A combination of words and sentences seems to work, as well as just an enumeration of words.
Thanks for the post!
Glad you like it! My goal is to get new styles/concepts for Flux. PonyXL has a huge library of styles and naturally understands concepts so I want to be able to replicate that in Flux.
LOL yeah I've noticed that sometimes enumeration of words are fine but other times sentences produce better results.
CATalyst was captioned with anime/booru tags so that helps Flux work with them but I think its not strong enough yet.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.