Read below for pony instructions!

NoobAi V-pred version:

This version uses the same dataset as the one I posted for Pony previously.

This was trained on 395 high quality images from the artist Gintoai. Using batch size 2, gradient accumulation 4, I trained it for a total of 10 epochs (total steps unbatched would be 7,120 steps).

Trigger word is: g1nt0

The base model I trained it on was not NoobAi V-pred but rather a model based on it called NoobAI v-pred 1.0 with EQ-VAE, which I highly recommend as the lora performs better with this model.

Now as for the images, I posted a pair of images with the same exact prompt except the model used on it. On the left, I used EQ-VAE and on the right is the normal NoobAi V-Pred model.

No post-processing or adetailer/facedetailer has been used on any of the images here, however I do recommend at least running adetailer/facedetailer to fix up the faces.
Hires.fix is optional but will definitely help with details.

Settings I used for the images above:

34 steps (I recommend 30 max)

CFG 7
Euler ancestral - Sampler
SGM Uniform - Scheduler





~~~~~~~~~~~~~~~Pony~~~~~~~~~~~~~~


Style trained on the artist Gintoai's pictures.

Recommended weight 0.8 ~ 1.00

No trigger word needed for pony