👑 Krea2 DMerge is LIVE! Version 1.0 Debut! 👑
🚨 FIRST ATTEMPT & IT (kinda) SLAPS. Prepare to elevate your art! 🚨
My maiden effort merging into Krea2 has been rigorously stress-tested, and the results are undeniable. This checkpoint delivers 🔥 pure, high-fidelity performance right out of the gate!
🚀 WHY YOU NEED THIS CHECKPOINT 🚀
🖼️ LORA-FREE POWER: The core weights shine brilliantly on their own—no need to overload the system just to look good.
🛡️ ULTIMATE STABILITY: Got that specific scene or pose you can't get without a lora?This checkpoint has been tested across numerous external Loras without a single hiccup.
⚡ BLAZING SPEED: Optimized with INT8 CONVROT for lightning-fast generation, seamlessly integrating with ComfyUI's default loader.
Jump into the generator or ComfyUI, load the weights, and generate perfection. Drop a comment or add a post to show us what masterpiece you created! 🌟
See the visuals in the gallery! 👇
Added BF16 by request for those with more VRAM or who want that extra punch.
🙏 Huge thanks to the creators whose incredible work made this merge possible:
This merge simply wouldn't exist without the fantastic models from @GBRX and @lonecatone23 . Both are genuinely excellent creators, and their models are the foundation everything here is built on — I'd encourage everyone downloading this to go check out their original work and show them some love. Standing on the shoulders of giants. 💙
Description
BF16 version 1
FAQ
Comments (15)
Wow, those faces look familiar. I'm sure @GBRX would agree.
It's common practice to credit the people whose models you merged as well as check permissions. Just FYI.
Permission between pirates is hilarious. I'm sure the other guy asked permission for those faces from the ladies, signed contracts, and brought their moms flowers in case the anal poses were too much
@agnesthecakeslayer You're not too bright, are you?
@lonecatone23 when you need to resort to petty insults, you've already lost the argument
@agnesthecakeslayer what arguement? You posted for attention. I pointed out a fact. Not much to argue about 🤔
you do see that gbrx left a comment before you did right? its literally two over to the right and one up. doesn't seem too upset. almost as if they knew about it but idk. maybe it's a different gbrx and the author gave you and gbrx credit. if you hit the "see more button" soooo what fact are you pointing out 😂
@053nodnarb503 GBRX left a comment after me as I told him. Oddly enough, he also used mine to merge and I have merge permissions turned off. I could have the model yanked.
He also added that part after. I made the comment.
Best Krea2 model. Thanks Bro.
Great job on this merge! I'm happy that you decided to include my model as one of the ingredients. Like chocolate and peanut butter, sometimes you mix things together and you get something both different and wonderful!
WoW. Did test a whole bunch now... THIS one is great! Very good mix of realism and creativity!
When using the int8 version. The skin often looks dirty, having dirt in small squares all over. Sometimes the overall image looks pixelated or like there is a hex grid in areas. Is this normal? What would I do to fix this, just use an upscaler? I don't have this issue with Bf16, but I am using an 8gb video card.
I haven't tested on an 8GB card but almost all my samples are done with the INT8 and I didn't have this issue. Are you using the normal comfyui model loader? Any loras or other things?
@daceheg192491 Sorry for the late reply, I didn't see the message. I am using the RLY KREA 2 Basic Edition template (downloaded from his Discord). I have tried varies Qwen Loaders. I have tried with and without Loras. I also have the same issue, but to a lesser extent with the Bf16 model. I know that it is on my end. * Standard settings CFG 1, Ksampler res_2 (Have tried multiple others), Basic Scheduler (Beta, 4 steps, denoise 1).*I am still not that adept at using Comfyui Desktop, so part of it is a learning curve on my end.
does this model work on forge?
I haven't tried it but as far as I know either of these should work. According to Gemini, Forge added INT8_convrot support in v2.27. The BF16 should load like any other Krea2 model.