This is my first ever consistent-character textual inversion, I followed the handy guide produced by BelieveDiffusion, creator of the LastName series. you can find the guide HERE
For similar consistent-character embeddings by other authors check out znobody, nobody and noexist too.
For most of the example images I've used Deliberate v2. I ran this against 9 different checkpoints and the results ain't too shabby, try it! testing results below.
Instructions: Place the .pt file under stable diffusion/embeddings and then use the prompt msred_deliberatev2 - the embedding is named as such because that is the checkpoint I trained it on, but it works for other checkpoints as shown above.
Note: I could have just included all perfect images in the examples, but that's not representative. I've included some not-so-good-outputs for completeness, some are blurry/lacking fidelity because I'm still working on my upscaling process (please post tips if you have any! much appreciated)
Oh, and beware the spiderhands and ten-toes. 😉
These were all trained and generated on a 12gb RTX 3060, results were speedy and flawless for a fairly budget GPU 👍
Checkpoints tested against:
SD 1.5
Credit to the authors of the above for some smashing models.
If I do any more of these I will tag them under dtmiss
Disclaimer: this is a fictional character and a experiment, any resemblance to actual persons is entirely coincidental.