World Furry Fusion – v1.0 (Refined)
📝Description
Checkpoint was created to generate images closer to the Loras (mainly Style Loras) using more precise prompts, resulting in images with more vivid colors and fewer problems, especially regarding the body structure of characters. The blending process uses the following base models:
The intention of the merge is to further approximate Lora's styles without losing its essence, such as "sketch images," "flat colors," "monochrome and grayscale," etc. The main highlight is the Illustrious version 2.0, which better captures the small details of the characters, in conjunction with a large part of NoobAI version 1.0, which includes a large database of trained images from the E621.
⚙️Recommendations
For more technical details about these versions, I recommend reading the articles at the link below:
Illustrious
https://civitai.com/models/1489531?modelVersionId=1684946
NoobAI
https://huggingface.co/Laxhar/noobai-XL-1.0
https://civitai.com/models/833294?modelVersionId=1022833
The model already includes VAE
Sampler - Euler A or DPM ++ 2M Karras (Use as you prefer)
Scheduler - Normal (ComfyUI / A1111 / Forge)
Steps - 30 steps
CFG Scale - 4 ~ 7
Hires. Fix / Adetailer / FaceFix (Recommended for best results)
Upscaler (Recommended for best results)
Positive prompt
masterpiece, best quality, amazing quality, very aesthetic, high resolution, ultra-detailed, absurdres, newest, scenery, depth of field, volumetric lighting,
negative prompt
worst quality, bad quality, low quality, lowres, signature, username, logo, jpeg artifacts, signature, watermark, artist name
The settings may change depending on the use of Loras, configurations, or where Checkpoint is running (Civitai or Local Generation), but all images in the Checkpoint gallery were created based on Civitai settings.
🤝Credits
We cannot forget this great creator of model.
-----------------------------------------------------------------------------------------
World Furry Fusion – v1.0 (Standard)
-----------------------------------------------------------------------------------------
📝Description
Checkpoint is my first and newest creation to generate new results (mainly furry characters) through more responsive prompts, and it has yielded quite satisfactory results. Checkpoint currently combines two of the main models currently used in Civitai:
The intention is to combine, through fusion, a set of images geared towards the "anime" style with a good dose of "furry". Therefore, don't limit yourself to just that; you can produce SFW and NSFW content, as well as humanoid characters, humans, famous cartoon characters, and more, whatever your imagination allows.
During post-training (merge), I noticed that there were a few artifacts during the generation of some images (this bothered me a lot), so I decided to add a VAE to add more colors and remove any artifacts from the images. I chose this VAE (Illustrious-XL-v2.0) because I am more familiar with it. If you have any questions about what a VAE is, I suggest reading this article:
-----------------------------------------------------------------------------------------
⚙️Recommendations
Sampler - Euler A or DPM ++ 2M Karras (Use as you prefer)
Steps - 30 steps
CFG Scale - 4 ~ 7
Hires. Fix / Adetailer (Recommended for best results)
Upscaler (Recommended for best results)
Positive prompt
masterpiece, best quality, amazing quality, very aesthetic, high resolution, ultra-detailed, absurdres, newest, scenery, depth of field, scenery, volumetric lighting,
negative prompt
bad quality, worst quality, worst detail, sketch, lowres, jpeg artifacts, signature, watermark, username, artist name
The settings may change depending on the use of Loras, configurations, or where Checkpoint is running (TensorArtor Local Generation), but all images in the Checkpoint gallery were created based on TensorArt settings.
-----------------------------------------------------------------------------------------
⚠️Note
I'm open to feedback on the model and plan to add new versions with different styles in the future, including my Loras created so far (I need to study this part a bit more to create a very flexible model with a unique and specific style).
-----------------------------------------------------------------------------------------
🤝Credits
We cannot forget these two great creators of models.