First test of the Krea2 Turbo INT4 model NEW 6.9 GB NVIDIA RTX 20 and later versions — small and fast on older graphics cards.
Model merger using a short script created by ChatGPT
template to use for the merge https://huggingface.co/Hippotes/Krea-2-Experiments/blob/main/krea2_turbo_int4-convrot-simple.safetensors
parametres generation
euler simple er sde simple __8 steps 1080*1440_1024*1536_972*1296_ 896*1152
Thanks to https://civarchive.com/user/PrincesOfDarkness for the buzz
INT8 models are compatible___ use forge neo classic
Preference parameters: ER SDE SIMPLE and ER FlowMatchEulerDiscrete, 8 to 12 steps
12 steps without LoRa and 8 steps with LoRa seems like a good compromise
This model mainly generates paintings or images, not ideal for realism and NFC content
Description
INT8 models are compatible
Preference parameters: SDE SIMPLE, 8 to 12 steps
10 steps without LoRa and 8 steps with LoRa seems like a good compromise
FAQ
Comments (4)
Will there be an FP8/BF16 version? Not everyone has Nvidia cards.
No, sorry, because the int4 and int8 models were created using a script written on ChatGPT specifically for these models
Good model, but it very like to generate paintings or images, but not a photo. For photo need to mention in prompt, that we need a photo. But model is good, dont making NSFW when it dont need. Keep this way.
You are right; it has been updated.