/vt/ - Virtual Youtubers

>>37126125
The results of the log isnt accurate to what the embed actually knows. The log just reads whatever promts are in the .txt file of the image it is currently looking at and applies whatever knowledge you have learned in the previous generations.(I dont think the output is produced by img2img but I idk)

It goes very quick., by the first few generations you should have something if done right.

The best test for the use case of what the embed actually do is to keep it as simple as possible, I do my quality tags, embed name, 1girl, standing.
Because a lot of the art it is trained at are portraits of the face it will help shows the weaknesses of the embed and any corrections you can do for prompts.

You can also test every embed it spits out during training too and compare alongside eachother. Takes a while to render but it is usefull for debugging or selecting a previous step as the final one instead.