[cite_start]There are three different training processes, each with its own job ID, start time, and base model[cite: 1, 2, 3]. [cite_start]All three training jobs have the same number of files and labels (75 each) and use "tag" as the label type[cite: 1, 2, 3]. [cite_start]They all use the Kohya engine, a resolution of 1024, and have a clipSkip value of 1[cite: 1, 2, 3]. [cite_start]The loraType is "lora" for all of them, and maxTrainEpochs is set to 19[cite: 1, 2, 3]. [cite_start]All three jobs also have a targetSteps of 9975 and have enableBucket, shuffleCaption, and flipAugmentation enabled[cite: 1, 2, 3].

***

### Job 563759-1

[cite_start]This training job started on September 8, 2025, at 6:55:14 AM and used the Pony base model[cite: 1]. [cite_start]The optimizerType is Prodigy, and the lrScheduler is cosine[cite: 1]. [cite_start]The networkDim is 64, and the networkAlpha is 32[cite: 1]. [cite_start]The trainBatchSize is 5, and numRepeats is 35[cite: 1]. [cite_start]It also has a noiseOffset of 0.03 and a minSnrGamma of 0[cite: 1].

***

### Job 563759-2

[cite_start]This job began on September 8, 2025, at 6:56:09 AM, with SDXL as the base model[cite: 2]. [cite_start]It uses Adafactor as the optimizerType and cosine_with_restarts for the lrScheduler[cite: 2]. [cite_start]The networkDim is 32, and the networkAlpha is 16[cite: 2]. [cite_start]The trainBatchSize is 4, and numRepeats is 28[cite: 2]. [cite_start]This job has a noiseOffset of 0.1 and a minSnrGamma of 5[cite: 2].

***

### Job 563759-3

[cite_start]Starting on September 8, 2025, at 6:54:11 AM, this training used the Illustrious base model[cite: 3]. [cite_start]Similar to the SDXL job, it uses Adafactor as the optimizerType and cosine_with_restarts as the lrScheduler[cite: 3]. [cite_start]The networkDim is 32, and the networkAlpha is 16[cite: 3]. [cite_start]The trainBatchSize is 4, and numRepeats is 28[cite: 3]. [cite_start]It also has a noiseOffset of 0.1 and a minSnrGamma of 5[cite: 3].