<h3 id="introduction">Introduction</h3>Krea2 Cat Tower is is <a target="_blank" rel="ugc" href="https://huggingface.co/krea/Krea-2-Raw">Krea 2</a> fine-tuned model.This model was trained with the goal of enhancing anime style.Hugging Face page is here:<a target="_blank" rel="ugc" href="https://huggingface.co/nukomasshigura/Krea2-Cat-Tower">https://huggingface.co/nukomasshigura/Krea2-Cat-Tower</a><h3 id="how-to-get-started-with-the-model">How to Get Started with the Model</h3>In order to perform inference, the following files are required in addition to this model.<ul><li>Text Encoders (use one of them)<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea-2/blob/main/text_encoders/qwen3vl_4b_bf16.safetensors">qwen3vl_4b_bf16.safetensors</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea-2/blob/main/text_encoders/qwen3vl_4b_fp8_scaled.safetensors">qwen3vl_4b_fp8_scaled.safetensors</a></li></ul></li><li>VAE<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea-2/blob/main/vae/qwen_image_vae.safetensors">qwen_image_vae.safetensors</a></li></ul></li><li>Turbo LoRA<ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea-2/blob/main/loras/krea2_turbo_lora_rank_64_bf16.safetensors">krea2_turbo_lora_rank_64_bf16.safetensors</a></li></ul></li></ul>You need to use the webui that supports Krea 2 and INT8 ConvRot.<ul><li>ComfyUI: v0.27.0 or later (v0.28.0 or later is required to use the INT4 ConvRot version.) <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea-2">Refer here for how to place files in ComfyUI.</a></li><li>Forge Neo: 2.27 or later</li></ul><h3 id="quantized-and-bf16-models">Quantized and BF16 models</h3><s>Only the INT8 ConvRot model is available on Civitai.</s><s>If you need other quantization models such as INT4 ConvRot, FP8 or NVFP4, please download them from the </s><a target="_blank" rel="ugc" href="https://huggingface.co/nukomasshigura/Krea2-Cat-Tower"><s>Hugging Face page</s></a><s>.</s><s>v2.0 BF16 model are also available on the Hugging Face page.</s>All Quantized models and BF16 model are available on Civitai."int8" (INT8 ConvRot) is recommend for inference.<h3 id="recommended-setting">Recommended Setting</h3>It adheres to the official settings of Krea 2 Turbo.v2.0 is Raw model, so I recommend using Turbo LoRA for inference.It adheres to the official settings of Krea 2 Turbo.<ul><li>Resolution: 1024-2048px</li><li>Steps: 8</li><li>CFG scale: 1</li><li>Sampler: euler, er_sde</li><li>Scheduler: simple, normal</li></ul><h3 id="prompting-guidelines">Prompting guidelines</h3><a target="_blank" rel="ugc" href="https://github.com/krea-ai/krea-2/blob/main/docs/prompting.md">Please refer to the Krea 2 Prompting guidelines.</a>I recommend constructing your prompt using the following structure. Main Subject + Pose/Action→ Appearance, Clothing &amp; Details→ Props &amp; Materials→ Composition, Framing &amp; Perspective→ Environment &amp; Background→ Lighting, Color Palette &amp; Mood→ Overall Aesthetic &amp; MediumUnless you specify a particular style in the prompt, the default style will be anime-style.<h3 id="training">Training</h3><ul><li>Trained using a dataset of images generated with <a target="_blank" rel="ugc" href="https://civarchive.com/models/920709/cat-tower-noobai-xl-checkpoint">Cat Tower</a> and a dataset of images not generated by AI.</li><li>100% Natural-language, no booru-tags</li></ul><h3 id="limitations">Limitations</h3><ul><li>Sometimes, sweat droplets may appear on the character's face even if you haven't specified it in the prompt.</li></ul><h3 id="license">License</h3><a target="_blank" rel="ugc" href="https://www.krea.ai/krea-2-licensing">Krea 2 Community License</a>

Krea2 Cat Tower

This is a Raw model. I recommend using <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/Krea-2/blob/main/loras/krea2_turbo_lora_rank_64_bf16.safetensors">Turbo LoRA</a> for inference.1) Datasets have been refined<ul><li>The captioning method was changed from JoyCaption to Qwen 3.8/3.6 or Gemma 4. Additionally, all captions for the main dataset were manually checked and corrected.</li><li>Replaced some training images and added a small number of new training images.</li></ul>2) Changed the fine-tuning method from LoRA merging to full fine-tuning using Fizgig.3) Changed the size of training buckets from 1.0 MP to 1.5 MP.Current issues<ul><li>Sometimes, sweat droplets may appear on the character's face even if you haven't specified it in the prompt.</li></ul>

Files with SHA256: b166ed401a0dba184f576e1b3a99b901abaa1ea18640eceec71eb06cced8496a