Description
Embedding trained for 2000 steps on ‘v1-5-pruned-emaonly’ checkpoint, so it should works well with most of v1.5 forks and mixes. I used dataset of 60 images with 680×680 resolution, and learning rate: “0.05:10, 0.02:20, 0.01:60, 0.005:200, 0.002:500, 0.001:1200, 0.005”.
!!!ATTENTION!!!
Some checkpoints have a lack of training on afro people. For example with PirsusEpicRealism checkpoint Emmanuella sometimes lost black skin with “green dress” in prompt. If you get issues like that try to simplify prompt or change checkpoint.
Also AIDA_EmSam embedding is incompatible with all LoRAs, trained on mostly light skinned European or Asian people. For example Realistic Glossy Face Skin LoRa RSERomantic_RSESofiko_RSEEmma-v1 could simply replace Emmanuella with any random European girl.
Preview images were rendered on checkpoint: realisticVisionV30_v30VAE (hash: c52892e92a)
If you are going to remake preview images you should have:
Checkpoint: realisticVisionV30_v30VAE
Embeddings: BadDream, FastNegativeV2, UnrealisticDream, badhandv4, kkw-Extreme-Neg, ng_deepnegative_v1_75t, kkw-ph1, AS-Young-neg, kkw-old, AIDA_EmSam.
Download link to AIDA_EmSam embedding will be available for you after purchase. For your comfort I packed AIDA_EmSam.pt into .zip archive with AIDA_EmSam.preview.png and AIDA_EmSam.txt , preview and description are compatible with AUTOMATIC1111 extra networks menu.
Also you will get .zip archive with all preview images and generation data for them.
Purchasing this asset you will achieve rights for selling images generated with AIDA_EmSam embedding. If you want say special thanks to me, you can add additional amount to your payment.
Please contact me if you need some extended rights for sharing or selling AIDA_EmSam, merging AIDA_EmSam with other embedding, using AIDA_EmSam in generation services and so on.
Reviews
There are no reviews yet.