--Anima
Trained on anima3. Recommended strength: 0.6.
Works best with block filtering — I created a ComfyUI nodepack for this: ComfyUI-LoRA-Block-Filter. Examples generated alongside my other LoRA — nicegirls.
Experimental version. Results may vary depending on prompt and block weight settings.
--Flux2
I use it with res2m+beta57, guidance 4, and it works very well. The quality is similar to my Qwen Lenovo, but I noticed that Flux2 is more flexible and has better prompt adherence.
Recommend to use with strength 0.85-0.9
-- Z-image Turbo
Actually i'm not sure that my workflow is correct, i believe the result could be better
Recommend to use with strength 0.80-0.9
I used euler + sgm_uniform with 4 guidance
P.S: at this moment for ComfyUI u need this pack to load lora properly https://github.com/PGCRT/CRT-Nodes
--Chroma
My first try to train lora that enhance realism in Chroma.
I'll upload my workflow today a lil bit later (or atm u can just take it from images) cause there some nuances about settings. But from my observations i noticed that for fast gen dpmpp2m + beta with cfg 3 works good . More experimental settigns are res_2s_rkmk2e + beta57 and cfg 3, more accurate gens, better limbs and fingers, but gives a bit weird grain on very small details + plus this method is pretty long
--Qwen
Works perfectly and much better then other versions. Gives extremely "live" images with amateurish quality with dramatic light and shadows, some blur and with vibe of 2000s street photography
You can take workflow here https://huggingface.co/Danrisi/Lenovo_Qwen/resolve/main/Qwen_danrisi.json
Thanx to fox23vang226. i understood that it's possible (and even have better effect) to use lora with values 1.5 - 2.0.
P.S.: almost all example images generate with both loras (NiceGirls)
--Wan
This is the first release — mainly a test to explore Wan's capabilities.
Note: In this initial version, facial details are somewhat lacking. For better results, I recommend using it alongside the Instagirls LoRA by my friend, which significantly enhances facial features.
--Flux
Who needs a fancy name when the shadows and highlights do all the talking? This experimental LoRA is the scrappy cousin of my Samsung one—same punchy light-and-shadow mojo, but trained on a chaotic mix of pics from my ancient phones (so no Samsung for now)
Description
FAQ
Comments (44)
All of my images have a really weird look to it. It's hard to describe but there are really weird artifacts in the image. I tried everything from using your workflow to updating comfyui. Nothings works. Any Idea?
What about the prompt? Also, what Torch version are you using? (Yeah, I remember it had a big impact on generations, even with the same prompt and seed, results could vary wildly between Torch versions.)
Danrisi I'm using 2.6.0+cu126. I just tried your exact prompt from an example image. I just updated comfyui and all dependencies. Just in case. I will try a newer torch version.
Gamblerschoice if u have windows then it's the maximum for you, as i remember there are some issues with 2.6+
I also have 2.6.0+cu126
Danrisi i m not having good result with 8 step lora ,does that work with that?
markhassain3712 No wonder — even at 30 steps (without LoRA, I mean) I noticed a loss of details compared to 50 steps
Danrisi I downgraded to 2.5.1+cu121, unfortunately images still look extremely weird. A friend of mine has the exact same problem, so it's definetely not my system. Does GPU Driver affect it?
Danrisi ohh ok
Gamblerschoice Have you tried different schedulers?
yoloswagg45 Yes just did... no difference. (they definetely do something different, but the images still have the same weird artifacts)
I think I just fixed it... for some reason I was using a distilled model. The images still look weird, but thats because of settings. But there are no artifacts anymore. But thanks a lot for your help! :)
@Gamblerschoice which settings do you made for fixing it?
@wopawopa01733 Ah I can't remember... it was mostly the wrong checkpoint (the distilled one). I think I just copied the rest from other posts.
Does this have to be used with a GGUF checkpoint? I'm trying to use it with the FP8 checkpoint and it doesn't seem to work at all.
Any thoughts?
I have the exact same issue, OP says it should work but I don't think it does.
Pretty sure it wouldnt work with fp8 .. Its a tiny slice with Qwen and loses a LOT of content... Try GGUF works a treat!
Excellent. I like it at about .7 and it works well with 10 or more steps. Very happy with this.
If you have a RTX 4000/5000 card with 16GB of RAM or more you can run the the model above (even though when you download it it is 20GB) easily as fast as a GGUF Q5 model that's around 13GB. Please test this out as you will get better result out of your images.
Also this: https://huggingface.co/nunchaku-tech/nunchaku-qwen-image
so this LoRA works both WAN & QWEN?
Yes. I have 3 different versions. For flux, qwen and wan
Dope, love it.
Always pushing the limits. Always keeping up with the latest LLM. A must have loras for any creators.
Appreciate that. Gotta keep the bar high, right? More stuff’s coming soon too 🔥
Danrisi Facts . Made me go straight to YouTube for how to for qwen . 🔥🔥
We are waiting for an UltraReal finetune of qwen, that can produce both amateurish and professional dslr like photos. Qwen image is not very good at realism as a base model, but it has "exceptionally well" response to training. BTW I liked your sony flux lora a lot, it was pure cinema. Just hope your qwen finetune would also encompass that kind also
May I use this Lora to generate videos for the Wan22 model? Should we connect this Lora to a low noise model or to a high noise model at the same time?
I didn't try but should work i think. You should connect to low noise only i guess
The Lora for Qwen does not load correctly. I tried the Lora loader, lora loader only & Lora Loader Stack (rgthree) and I got : lora key not loaded: [...]`
Is there a sometihng I didn't get ? I also did reuse avalaible workflow on the examples. The Lora weight around 45mb.
Hi. lora key not loaded: [...]` means that u need to update ComfyUI
Looks really good!
Legend fr
Possible to use this in wan2.2?
Yeah, you can use wan version for wan2.2
The qwen version of this lora slaps so hard its unbelievable. I don't know if it was your prompts or what but this model can do some serious heavy lifting. Nothing is getting close to it
Thanx a lot <3
Could you make more Lorasfor wan2.2, its really great for generating images
Okay. It's also in my plans
Has anyone figured out how to run this with the lightning 8 or 4 step adapters? Doing so seems to completely kill the lora. Is everyone running this at 50 steps?
On 30 steps it looks okay too. But yes, lightning lora kills quality and it looks like SD1.5 or dalle2
It's working fine for me with 4 steps v2, lora at 1.5, in the order 4 steps first, then lonovo.
This works great for Qwen. Thanks.
"Just when I thought I was out, they pull me back in!"
Glad that u like my model so much. hehe_boi.gif
thank you x