You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
What's the point: Pony-based models are quite unruly, and often only give me what I need after a long series of generations. In the process, I discovered that there is a behavioral pattern in which the model generates 2-5 images in a row that are not very different from each other. Then the influence of tokens somehow changes and a number of similar generations follow, but with the predominance of some other specific token. Thus it turns out that one query can get several conceptually different results. But for each such result there will be several slightly different images (as if I used Vary (Subtle)).
That's why when it's time to experiment I'd like to set the step for a seed to be greater than 1, in order to skip similar generations.
What's the point: Pony-based models are quite unruly, and often only give me what I need after a long series of generations. In the process, I discovered that there is a behavioral pattern in which the model generates 2-5 images in a row that are not very different from each other. Then the influence of tokens somehow changes and a number of similar generations follow, but with the predominance of some other specific token. Thus it turns out that one query can get several conceptually different results. But for each such result there will be several slightly different images (as if I used Vary (Subtle)).
That's why when it's time to experiment I'd like to set the step for a seed to be greater than 1, in order to skip similar generations.
Originally posted by @DavidKouw in #2597 (reply in thread)
The text was updated successfully, but these errors were encountered: