Introducing 10 Toes Raw V1.3
This is the RAW version of Krea 2 with my 10 Toes V1.3 finetune. I prefer using the turbo LoRA with the RAW model to generate less steps.
Here's the Turbo LoRA: loras/krea2_turbo_lora_rank_64_bf16.safetensors · Comfy-Org/Krea-2 at main
I use it with a strength of 0.8-1.0. Workflow is the same as Turbo: CFG: 1-1.2, 8-10 steps, Euler/Euler Ancestral or Er_SDE, Bongtangent or Beta75.This model will do NSFW and SFW, but with a stronger adherence to NSFW. The Krea 2 model has strong safety features built in, but it's also built on strong anatomy and NSFW content. It's a bible of filth to be honest with you.
In my experience Krea 2 RAW works with any Turbo LoRAs - kif-kif. This finetune plays nice with all LoRAs. Just treat it like a Turbo model.
🚨Important🚨:
There are two convrot models for you to download: INT8 and INT4. The savings of using them are massive, no matter your GPU. For the fastest speeds, use INT4. For the highest quality and very fast speed, use INT8.
What the fuck is "INT8 and INT4 Convrot" bro?
Good question - it's an optimized INT8 and INT4 quantization that let's you goon hard and fast in ComfyUI. If you're running the nightly build of Comfy, with Pytorch 2.10 or higher, Python 3.12 or higher and cu130 or higher, then you're golden. It adds significant speeds to your renders. I've also uploaded on Huggingface the text encoders in INT8 Convrot too!
Want the Turbo finetune? No problem, grab it here: Winnougan 10 Toes (INT8 Convrot & FP8) - v1.3 | Krea 2 Checkpoint | Civitai
Consider supporting me on Ko-fi. Even $3 goes a long way!
I'm burning through resources, electricity bills and shaking with fear if anything should happen to my ram sticks or SSDs in this alternate reality where ram costs more than a G-Wagon.
👇🏻👇🏾Support me below 👇🏿👇🏼
💚 Workflow: (https://pastebin.com/nHWpSccV)
💙 Support on Ko-fi: (https://ko-fi.com/winnougan)
I'm saving toward a hardware upgrade so I can expand into Ideogram 4 and LTX-2.3 video LoRAs and deliver releases faster. Every coffee helps make it happen. ☕
All LoRAs have the last 4-8 epochs available for members on Ko-fi along with datasets and sample training images.
🖤 YouTube — Tutorials and reviews: https://www.youtube.com/@Winnougan
💀 Discord — Releases, requests, and conversation: https://discord.gg/CJv5wceJaN
Description
FAQ
Comments (11)
Thanks for the raw. This model gets me very close to raw+snofs+style loras, just as long as I use a filter_bypass -- which is alright, too. Might actually try this for a few days and see if I even notice its on. With the bypass it's pretty much passed the fidelity of my original raw outputs.
I found that you don't need the filter bypass - but it's a 3kb file, so why not. I think I might bake in the filter bypass for V1.4
@Winnougan Only the penis-heavy prompts needed the bypass, at least when the penis was not the focus of the image (for example, handjobs at a distance). I'd only recommend against doing it for the same reason for the raw to exist: to leave it as open to interpretation. There are like 5 or more different 1k bypasses, I've been using s1lverco1n's, but other's may work for other people. My only negative is that the bypass can let in a lot of the anime styling. In my opinion it'd be best to round out knowledge and styles without depending on baking in the bypass (cause ideally I'd have it off for polluting my images with anime styles lol).
edit: does seem like SNOFs is still important to keep around for some prompts, too. But it's getting there. Feeling more like SDXL by the day lol
doesn't work in wan2gp. any advice? I replaced the Krea2 Raw finetune file
I only use Comfyui. Which quant
Got it working with wangp, but too much anime, even after prompting it out... odd
I found that the softness and roughness of the int8 are combined too much. Is it a problem with me?