Type | |
Stats | 716 |
Reviews | (64) |
Published | Nov 8, 2024 |
Base Model | |
Training | Steps: 12,000 Epochs: 3 |
Hash | AutoV2 0E6AF806B3 |
GREAT SCOTT! Vivid insanity is here!
===========================================================================
v2.beta - (11/21) - the first look at Schnell's v2 variant (s_v2_alpha) is out and ready for your enjoyment. Let us know what you think - we look forward to seeing what you make! =)
Note: Use LoRA with C4PACITOR at slightly lower than normal weights for best results and to avoid anatomy blending.
===========================================================================
v2 - 11/5 - Training recommences shortly after dataset and compute adjustments. Thanks for your patience!
v2.alpha - (10/20) - The first taste of v2 (Analog)'s full bf16 fine tune is here. Read model description for more details. Much more to come. Stay tuned.
v1.1 - (10/14) - Enjoy this re-balanced release of dev/schnell v1 while v2 is cooking. It behaves better and has been completely rebased - thank you sharing your feedback and creations!
v1 - (9/27) - FUX C4PACITOR updated!
v1 - (9/22) - s_v1 released! We'll be sharing some merges/goodies in the coming days. V2 training is in data selection/validation phases. Training starts soon!
(9/15) - d_v1 released! We look forward to seeing what you craft - please share your creations and feedback on this page! FUX C4PACITOR (LoRa pipeline embedded in the models) updated to contain RC version weights. We continue to release checkpoints first and the LORA from the last release on that page for those who generate on site or to use for their own purposes.
(9/8/24) - Early beta releases are here! S-models have both FP8/16 on a single tab (and will be released this way henceforth). Enjoy! Training/fine-tuning continues while gathering early feedback! Please share your generations!
v0/d_v0/s_v0 - A hint of things to come. These are the first of our Flux variants, to be released with the NSFW realism LORA behind its power in the coming days. S is fully Schnell based, D is fully Dev based.
Update 9/2/24 - FUX C4PACITOR (LoRa pipeline embedded in the models) released. Just over 200mb of v0 power to go! Enjoy!
C4PACITOR models are created by DR34MSC4PE with all the trappings you've come to expect as well as some cool bonuses:
Enhanced realism and photo-realistic concepts, trained on high quality datasets with the latest techniques.
Specific anime/illustration tuning to introduce some lost artistic concepts
NSFW tuned and capable of realism and artistic/anime images. Female anatomy is well represented with additional fine tuning in the works.
Triggers are strongest with "nsfw" and "woman".
Exceptional performance with character/other/stacking Lora
Limitations:
Nude male anatomy (WIP)
Hardcore NSFW concepts (WIP)
===========================================================================
DR34MSC4PE is
c0ur4ge (trainer/qa/inference code) /
eraser851 (captions/tooling code/data/qa)
===========================================================================
Recommendations (v1/v0/d_v0)
===========================================================================
Guidance : 3-7 (use variable values for NSFW ~4)
Steps: 18-60
Sampler: DEIS
========================================
(s_v0/s_v1):
===========================================================================
Note: 100% Schnell Base
Steps: 2-12
Sampler: DEIS
Note: Trigger words are stronger for Schnell's base to make up for the inability to control guidance.
===========================================================================
We look forward to updating the series frequently with new finetunes, LORA and other FLUX goodies. Please share your feedback and generations!