Well, somebody argued that training AI slop and merging models also requires "LOTS" of time and efforts, even hours of H100.
This ckpt is a speed-run demonstration of AI slop training and how to release ckpt as style LoRA.
Grab 89 AI slop n**w images from Pixiv with downloader extension, ~5min. Took some time handpicking good looking images instead of just auto grabbing.
Auto caption with LLM. ~1min.
Train a LoRA on the cheapest 4090. 1600 steps. ~12min. Probably overtrained.
Merge the LoRA into ckpt in ComfyUI. <1min.
How does the ckpt look? (hmmm, tbh looks not bad)
Took me ~20min, that's "LOTS" of time and efforts.
FYI:
Based on RDBT v2.4
16-step acc LoRA included. So use CFG 1 and step 16. Because AI style has no texture, there is nothing to loss, the distilled output is almost identical, but 4x faster.
Only uploaded int8 ckpt to save my server quota.
To determine if a model has undergone extensive training:
Don't rely on how strong its default style is. On the contrary, a strong default style indicates less training data because it was overfitted.
Instead, look at the breadth of knowledge the model possesses. For example, its unique responses to various completely different styles.
If someone claims their model has been extensively trained, even using H100, and that the model has a strong default AI slop style, they are simply lying.
[this_is_the_taste_of_a_liar.webp]
Sharing merges using this model is not allowed. This "restriction" won't affect anyone. It's only aimed at those who steal others' models to sell.
Special trigger words and latent watermark included.