Red Sea

CHECKPOINT
Reprint


Updated:

Red Sea is a furry SDXL lightning model designed to default to realistic/semi-realistic look, based on seaart lab's excellent furry model.

General usage:

Image sizes are the same as seaarts, ideally around 1024x1024, 1152x896, 896x1152 but variants should be fine.

It should default to realism/semi-realism however if there are issues the best tags to push this are : Photo, raw, realistic,

E621 tags work fine, seemingly with or without underscores.

Example pictures were done with auto1111 using high res fix to upscale by 1.5.

any sdxl VAE should be fine, I simply used the standard one for sdxl.

Samplers/steps :

This is a lighting model and uses very few steps/low cfg. the ones I tested with and work were:

Euler a with 8 steps and 2 cfg

DPM++ SDE Karras with 4 steps and 2 cfg.

Ideally you will need a 8gb vram card, you maybe able to use less with the fp8 features in auto1111/comfyui:

in Auto1111 this is done by going to settings > Optimizations > FP8 weight and setting this to enable for SDXL.

in comfyui adding --fp8_e4m3fn-text-enc --fp8_e4m3fn-unet as launch parameters.

Note this will lower the quality of the images.

Quirks:

Apart from usual anatomy issues poses can be a bit bland and might need clever prompting to get around.

Credits:

Gungie, who posted this model on Civitai

Seaarts Furry SDXL model - If you enjoyed this model please remember to upvote this one too: https://tensor.art/models/714704149125790858 from https://civitai.com/models/391781/seaart-furry-xl-10

A_B - A wonderful friend to Gungie, who has consistently made SD more fun. ^^

Version Detail

SDXL 1.0
Red Sea is a furry SDXL lightning model designed to default to realistic/semi-realistic look, based on seaart lab's excellent furry model. General usage: Image sizes are the same as seaarts, ideally around 1024x1024, 1152x896, 896x1152 but variants should be fine. It should default to realism/semi-realism however if there are issues the best tags to push this are : Photo, raw, realistic, E621 tags work fine, seemingly with or without underscores. Example pictures were done with auto1111 using high res fix to upscale by 1.5. any sdxl VAE should be fine, I simply used the standard one for sdxl. Samplers/steps : This is a lighting model and uses very few steps/low cfg. the ones I tested with and work were: Euler a with 8 steps and 2 cfg DPM++ SDE Karras with 4 steps and 2 cfg. Ideally you will need a 8gb vram card, you maybe able to use less with the fp8 features in auto1111/comfyui: in Auto1111 this is done by going to settings > Optimizations > FP8 weight and setting this to enable for SDXL. in comfyui adding --fp8_e4m3fn-text-enc --fp8_e4m3fn-unet as launch parameters. Note this will lower the quality of the images. Quirks: Apart from usual anatomy issues poses can be a bit bland and might need clever prompting to get around. Credits: Gungie, who posted this model on Civitai Seaarts Furry SDXL model - If you enjoyed this model please remember to upvote this one too: https://tensor.art/models/714704149125790858 from https://civitai.com/models/391781/seaart-furry-xl-10 A_B - A wonderful friend to Gungie, who has consistently made SD more fun. ^^ License: CreativeML Open RAIL++-M: https://github.com/Stability-AI/generative-models/blob/main/model_licenses/LICENSE-SDXL1.0

Project Permissions

Model reprinted from : https://civitai.com/models/402017/red-sea?modelVersionId=448263

Reprinted models are for communication and learning purposes only, not for commercial use. Original authors can contact us to transfer the models through our Discord channel --- #claim-models.

Related Posts