Loading
Vous avez un workflow ComfyUI ? Obtenez une documentation lisible directement dans votre navigateur.Rendre un workflow lisible.Essayer maintenant ↗
Trained on a dataset of 55 handpicked images.
Source du modèle
Description de la source
Trained on a dataset of 55 handpicked images.
As always, I recommend euler + beta57
Krea 2 Version
Trigger word: sgca style
The bf16 variant produces the same quality output as fp32. However, there are minor differences in details. You have to play Where's Waldo to spot them, though. That being said, I personally prefer the fp32 variant after testing with side-by-side comparisons.
The bottom line: choose bf16 if you prefer smaller files. Choose fp32 if you've got the VRAM to spare.
Anima Version
Trigger word: @sgca style
Sources
1 sourceArtefacts du modèle
2 artefactssupergiant_games_concept_art_style_v1_kr2_bf16.safetensors
safetensors · BF16 · 224 MB · SHA-256 6ea4beafe93d…7a43 · Civitai
Téléchargersupergiant_games_concept_art_style_v1_kr2.safetensors
safetensors · FP32 · 448 MB · SHA-256 3b2f39489e0d…2a09 · Civitai
TéléchargerExtraits de sources
1 extraitSource context: 396 downloads · Type LORA · Base model Krea 2
Again, euler_ancestral fried all of my gens. I have no idea why. A quirk of T-LoRA, maybe?
