Lox's Utopic World 🫧 | Krea 2 V1.0 Quants

Lox's Utopic World 🫧 | Krea 2 - V1.0 Quants

NSFW

🫧 UTOPIC WORLD INTRODUCTION 🫧

🫧 UTOPIC WORLD 🫧 is my first Krea 2 model ! Made with passion for arts, it will helps you to bring creation at its finest.

🫧 Two versions available ; BF16 and FP8 (~0.5%-1% loss compared to bf16). Use the version recommended to your hardware.

🫧 All showcase pics are made using BF16 version.

🫧 UTOPIC WORLD SETTINGS 🫧

🫧 For best results, I recommend the following settings :

Sampler : Er SDE
Scheduler : Simple
Steps : 8-12
CFG : 1

---> Also check my workflow

🫧 UTOPIC WORLD 🫧 by DrLox 💋

Description

V1 but quantized in :

INT8 Convrot
NVFP4

FAQ

Comments (17)

Birlalo_AIAug 10, 2026· 3 reactions

CivitAI

really nice model with very strong prompt adherence. this model is underated need to be recognised.

DrLox

Author

Aug 10, 2026· 1 reaction

Thank you ! It's more me that tends to generate curvy women haha

Birlalo_AIAug 10, 2026· 1 reaction

@DrLox same i had to push prompt to generate curvy on other model and while trying this model had to readjust a bit. thank you for this model, need a little more works but as of version 1 its really fun. hoping to get more upgraded versions.

DrLox

Author

Aug 10, 2026

@AI_Aurora What could be upgraded in your opinion ?

Birlalo_AIAug 10, 2026· 1 reaction

@DrLox my bad, as of now i don't think there's really anything that needs upgrade. since the prompting for this model is really strong so i had issues in earlier test with extreme outputs. had to relearn prompting specific for this model but after adjusting to new prompting the results are getting better. if i find anything i'll be sure to leave a feedback. ah. a detailed description is needed :)

DrLox

Author

Aug 10, 2026

@AI_Aurora Okay, what details should I add ?

Birlalo_AIAug 10, 2026

@DrLox well i just tried the fp8 model, there's definitely something not right in your int8 version also i use custom fast int8 (w8a8) node for other int8 finetunes which works really well but your int8 model doesn't seem to work on that node only on built-in node but the fp8 is giving a really better results. my guess is int8 conversion is not accurate.

DrLox

Author

Aug 10, 2026

@AI_Aurora Can you link the node pls ?

Birlalo_AIAug 10, 2026

@DrLox https://github.com/BobJohnson24/ComfyUI-INT8-Fast

DrLox

Author

Aug 10, 2026

@AI_Aurora Did you use it on others Krea 2 models ?

Birlalo_AIAug 11, 2026· 1 reaction

@DrLox yup works on all other int8 model

DrLox

Author

Aug 11, 2026

@AI_Aurora Can you share the workflow pls ?

Birlalo_AIAug 11, 2026

@DrLox cutom node int8 one's or the with normal load diffusion node?

DrLox

Author

Aug 11, 2026

@AI_Aurora All ^^

Birlalo_AIAug 11, 2026

@DrLox wait i will put notes instruction inside the workflow and upload it in my profile

ivar__the_bonelessAug 15, 2026

CivitAI

no int8 convrot ?

DrLox

Author

Aug 15, 2026· 1 reaction

It is in Quants version

Checkpoint

Krea 2

by DrLox

Download (Beta) View on CivitAI

Details

Downloads

295

Platform

CivitAI

Platform Status

Available

Created

8/10/2026

Updated

11/9/2026

Deleted

Files

qwen3vl_4b_fp8_scaled.safetensors

Size:

4.88 GB

SHA256:

54bd5144df0bbc25dd6ccadfcb826b521445a1b06ae5a42570bdd2974ca87094