This is where the VPRED converter loras and lycoris will go when a model is successfully converted.
Source du modèle
Description de la source
This is where the VPRED converter loras and lycoris will go when a model is successfully converted.
They are essentially TOOLS. Don't expect good results out of the gate. You WILL get good results if you use my models to merge and then train the merges.
BigAsp2:
In this case, the UNETS were not merged prior to training like the experts that I created for the feature extraction, I simply trained BigAsp2's UNET and froze the clips.
You may do whatever you want with these; I release with zero stipulations and zero rules.
The images for them probably mean nothing.
Stage 1 LOKR was 10k images 5 epochs center timesteps; meaning noise primarily.
Sources
1 sourceArtefacts du modèle
1 artefactasp2_omega_v1-1mil-finetune-vpred-000001.safetensors
safetensors · 757 MB · SHA-256 812bf0f6be92…2723 · Civitai
TéléchargerExtraits de sources
1 extraitSource context: 206 downloads · Type LoCon · Base model SDXL 1.0
Recipe;
BIGASP2 UNET
BIGASP2 CLIP_L + OmegaSim v73 CLIP_L
50/50
BIGASP2 CLIP_G
Stage 2 was 100k images (89k?) 4 epochs wider band timesteps, not quite fitted fully but close enough to make a merge with Omega. Omega being the grandmaster glue.
Recipe;
BIGASP2 UNET
Stage 1 LOKR
BIGASP2 CLIP_L + OmegaSim V73 CLIP_L
75/25
BIGASP2 CLIP_G