A primitive(?) merge of furry model with the anime.

fluffyrock-1088-zsnr-vpred-e255-minsnr-e29.safetensors / divineanimemix_V2.safetensors

Pros over furry model:

• Better understanding of anime LoRA;

• No fucked-up style no matter how many artists you put in;

• Kinda no artifacts in shaded areas.

Pros over anime model:

• Knows more concepts.

Minuses:

• Anime character LoRAs seem to work worse than on anime models;

• Mostly don't know anime model basic and acknowledgeable characters;

• Character detailing with rear view may be worse than front.

Installation:

1. Download and put the model (.safetensors) and config file (.yaml) to the stable diffusion models folder. A configuration file is required for the model to run;

2. Download CFGRescale from webui addons;

3. Download the tag file and throw it in the path "extensions\a1111-sd-webui-tagcomplete\tags" and activate it in the settings (you must have TagAutocomplete extension);

Extra QoL's:

1. Download Embed and LoRA:

   1. Negative embedding https://civarchive.com/models/87781?modelVersionId=94126;

   2. Quality tags LoRA https://civarchive.com/models/127533/fluffyrock-quality-tags;

2. FreeU - download in extensions (either by link or name):

   1. Start At Step - 0.1, Stop At Step - 0.45;

   2. The difference between "Version" and "Recommendations" seems to be at the margin of error;

3. Wildcards - download in extensions (either by link or by name), download the archive of cards for the model and unpack it to "extensions\sd-dynamic-prompts\wildcards". Contains cards by tag names, in which the authors who drew artwork with those tags are laid out;

4. LoRAs trained on furry models.

Generation tips:

1. Clip Skip 1;

2. CFG 6-10;

3. CFGRescale set to 0.6-0.7;

4. Render image size from 512 to 1088, but it is better not to set above 960 and not below 576;

5. In settings, set VAE encode and decode type to TAESD on both options, it will reduce overcooked look and prevent most of baked and burned out spots;

6. If you want humanoids - In the beginning of the prompt - "human, [not furry], your request", then add sex of the person: female or male or intersex, the latter is better to mix with female;

7. Add some artists to the request to get better result, example: by masha / by merunyaa / by cutesexyrobutts. You can find more artists here (add tags that you are interested in).

8. Prompting is similar to the danbooru (anime), and easily convertible to e621 prompt by using tag autocompletion file.

You can find more tips on the page of fluffyrock model - https://civarchive.com/models/92450/fluffyrock

Huge thanks to WithoutOrdinary for amazing and versatile model!

Huge thanks to TroubleDarkness for making anime-style model that is capable of good prompt understanding and good compatibility with LoRAs!

Another thanks to the torkroll from lodestone_rock's server for pointing out the merge recipe.

Please leave a review and describe what can be improved if you find an opportunity, or tell me if there any unlisted flaws compared to pure fluffyrock/anime models.

That can help others when reading this info.

Thank you for reviewing this model!