Foreword
In order to be able to estimate the influence of the number of training steps for future trainings of Embeddings, I created images at different training steps. The result can be seen below. There can be found images at 10000, 15000 and 20000 trainings steps. Parameters were for all images the same. The model is my WhiteMechanicalCat with additional Hires.fix Denoising Strength with 0.7 as factor.
Example Images
Figure 1: Image with 10000 Steps
Figure 1: Image with 15000 Steps
Figure 1: Image with 20000 Steps
Conclusion
At a certain point, the quality of the images seems to be good. It changes with more training steps what we get to see. As one can see, the results of the image creation are changing with the performed training steps significantly.
Accordingly, it makes sense to evaluate the results of image creation according to the training steps. A good model is one that delivers the desired result. Fewer training steps may result in even be the more desirable model.
Finally
Have a nice day! Have fun! Be inspired!