Version 1:

"ACR_nobake" or "ACR_nobake_float16"

This is a Merge between several checkpoints including, but not limited to:

Asterism - baked VAE | Stable Diffusion Checkpoint | Civitai

CyberRealistic - v3.3 | Stable Diffusion Checkpoint | Civitai

Rabbit - v1 | Stable Diffusion Checkpoint | Civitai

(Cyberrealistic_v33-A-small.safetensors)

(Rabbit_ver1_0.4W.vae.safetensors)

Not baked.

As for what it can do: Realistic anime or artistic anime art. It will not provide HDR photorealistic images - expect art similar to 3DG if prompted. You can keep it simple or have a beautiful landscape background - your imagination is the key.

It is also small enough it could be merged with other models.

Version 2:

Main Blending Factor With ACR (25%)

Specialized Cute Style (Lin)特化可爱风格(琳） - V9 | Stable Diffusion Checkpoint | Civitai

Style Blending Factor with ACR (5-15%)

Swoonify - v3.0 | Stable Diffusion Checkpoint | Civitai

guoz_mix1 - guoz_mix | Stable Diffusion Checkpoint | Civitai

Furnace 75v1 Cu6 cute style - v1.0 | Stable Diffusion Checkpoint | Civitai

E&D Manga (Style) - v1.0 | Stable Diffusion LyCORIS | Civitai

After Blending each with my main Model “ACR_no_bake” we tested each combination and chose the best blends to continue blending. The final blend is taken from numerous different combinations of each style with the Main Blending Factor, and with the main model and consist of both weighted sum and add-difference blending methods.

NOTE: About "no bake"

The final blend was only 20% new and 80% old, ACR_nobake. It fits within the similar way that ACR_no bake functions and while I cannot attest to the status of “no bake” for baked in VAE, it does fit within the model framework, so "no bake" in the title is only for classification purposes and not attesting to the status of the model.

Sidenote about negative prompts in regards to showcase images:

To get the same/similar picture you have to use the same prompts with the same seed. Although I no longer combine prompts in the traditional way, or by adding what I now consider to be nonsense, some of the images used for this model showcase for validating the model versus version 1 do use either the exact same, or as close as I could to the original version without introducing new artifacts.

When prompting for the first picture, which has hidden metadata, I had to add in a few negative prompts because the resulting image had significant changes in quality by introducing what I've dubbed "snow_flakes" in my new updated negative prompt for that and other pictures. Small white and scattered points in the image.

To preserve the mystery I have cleaned the metadata for the first image, but the prompts are visible for the other pictures.