--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 (48)
Pretty thick Loras! ❤️
For anyone who's wondering, this DOES work on WAN 2.2 low for image gen.
Yes. But i'll retrain next week for Wan2.2
@Danrisi Thank you, I'm looking forward to it! BTW, did you train this on image sets only? I've been contemplating training a Wan2.2 LoRA, but I only have image data.
@questionable_tastes u mean didn't I train it on videos also?
@Danrisi Correct. Basically, I was wondering about the feasibility of utilizing a Chroma dataset for Wan2.2 low training.
@questionable_tastes no, i didn't train on videos, but i have in plan to train style lora for wan2.2 (maybe soon on wan2.5 if it will be open source) with vhs/hi8 videos to gen with genuine analog effect
@Danrisi That's good to know! Didn't want to waste GPU power for trying to train a Wan LoRA on images if it wouldn't work.
For the Chrome one: Any workflow? I find Chrome to be hard to make a good workflow for. Also the 2K-QC is gone from CivitAI, only 2K-DC exist now.
Don't worry, i'll upload a few versions tomorrow maybe. At this moment u can take workflow under images i uploaded
@Danrisi how do you take the workflow from the images?
@halithaz Hi! Drag the image directly into a blank workflow, that should work for you!
@Danrisi thanks!
hi great job is the chroma version trained and compatible with nsfw?
Hi. Thanx. Chroma is support nsfw out of the box, therefore, I did not have to train additionally at the NSFW
Excellent LoRA, made me use Chroma (HD version) again and it produces great realistic results. Many thanks! Keep up the good work ...
Congrats on this excellent LoRA! Another masterpiece! I was surprised that the quality remains high even with the Flash LoRA. I have created a dedicated workflow for it: PhotoFlow Chroma HD Flash
Cool 😎
Thanx a lot, bro <3
I'll check workflow 🧐
It did not notify me that you had conjured more awesome!
It looks wonderful! Thank you!
Can you talk about Chroma for a second? They're real bad at explaining how to use their model. Did you need to go though any special here? I have seen a TON more slop than winners out of the model so I don't really get if it's a lora or model or workflow or settings problem.
It's a flux-s based de-distilled checkpoint. Most of the images generated on the web is not going to be good. You need to use comfyui or forge to get a good image.
The best way to prompt it is to find something similar to what you want (in style or content) online, then run that through joycaption, then use the joycaption result and modify it until you get what you need.
If you do that, it will create things that are frankly hard to distinguish from reality.
@brambola14 Great info many should know.
Your Chroma LoRA is the perfect foundation for those of us wanting to try Chroma because we can compare to Flux/Qwen/Wan because your LoRAs are THE standard for each. Thanks for these.
Thanx, glad to hear <3
Anyone using this on forge? What are your settings regarding distilled cfg and cfg?
I also cannot get it to work properly. I tried adjusting the diffusion in low bits, but that didn't change anything. Any help would be appreciated.
Use Sampler Deis and Schedule Beta, it works perfectly
Can't put my finger on it but Chroma models just don't seem that realistic to me... They seem airbrushed or lacking depth/proper lighting or something. Not sure how to get good prompts.
Hey! I’ve been using Chroma for quite a while, and I think I can share a few tips on how to get better results.
First off, if you’re not using ComfyUI but Forge or a similar Web_interface, there’s a chance you might need to change a few settings that are disabled by default. I personally use Comfy, so I can’t say exactly which ones.
However, I don’t want to self-promote too much, but I do have a Chroma guide on my CivitAI profile — it’s not perfect, but I tried to include the most useful tips that helped me achieve decent results.
One of the most common mistakes people make is Bad Scheduler +Sampler combo — it’s super important. Check what Danrisi uses for his LoRAs, and look at images from other users; you can often spot combinations that work well for them. I personally use Res_multistep + Beta.
Also, check out the comments under the main Chroma page on CivitAI — people have shared good advice there on how to get better results.
And of course, prompting matters a lot too — you might want to rethink how you structure your prompts. Sometimes just changing the phrasing makes a huge difference.
Love it! What settings did you use for Lora training on Chroma? And did you use kohya_ss?
Looks extreme low quality. Everybody uses it, while generates cgi pics, what for?
Pony 6 is more realistic.
Nah, my brother in Christ, I disagree. This is a very versatile model, it can do everything (almost), from realism to 2D art. It’s just a base model, designed to be easy to train. You can say whatever you want, but I’m genuinely having fun using it!
Bro, wtf
Yeah pony 6 couldn't make something realistic if it tried as it is a pretty bad base model but good to train on.
Hell, i can only rely on user pics and if this is ultrareal, you didnt use xl for a while. Just try new models, and your qwen, flux, chroma will look, like cartoon models.
@chamarrhoman377 Ok 👍
@Danrisi whatthefaka?
The thing is, new models are realistic only in motion with wan or refine with wan.
First of all — huge thanks to Danrisi for making this LoRA.
It perfectly captures the early 2000s vibe — I haven’t seen anything like it on Civit so far. Also, in my recent post — to demonstrate what Chroma can do — I shamelessly stole some generation ideas from Danrisi :^).UltraReal finetune it’s awesome!
I know this might not be the perfect place to post, but after reading some recent comments, I just wanted to share my thoughts — especially for those who are still unsure whether to try Chroma or not.
Hot take: Chroma doesn’t deserve the hate it’s getting.
It’s one of the most open, flexible, and community-driven models out there
You’ve been handed a new base model — with strong prompt adherence, no censorship, a solid license, and it’s literally built for easy LoRA and fine-tune training.
By spreading negativity without arguments, you’re not just hurting a few users — you’re hurting the open-source community as a whole.
We need projects like this to evolve.
Constructive criticism moves things forward — mindless hate does the opposite.
What’s the point of hating Chroma?
If you think SDXL is better — great, use SDXL!
Every model has its own strengths, so why not use both for what they do best?
“Looks extremely low quality” is not an argument.
Realism is a spectrum,
Just look at some examples — I think it’s a great baseline.
I’m not the best at prompting, but even with my simple setups, Chroma delivers solid results:
🔗 https://civitai.com/images/106836024
🔗 https://civitai.com/images/99916368
🔗 https://civitai.com/images/97558664
🔗 https://civitai.com/images/105539810
🔗 https://civitai.com/images/99916682
Juts remember chroma is a base model, not a finetune — yet it already delivers amazing stylized 2D and realistic results.
Not everbody talks about chroma. This is for every new models.
The closeup is good, you discovered sd 1.5, other ones are just ai.
@ananiajammal501 You’re right, I honestly thought this comment section was dedicated only to the Chroma LoRA.
I know I won’t change your mind, but honestly, prompting with tags like 1girl, 2boy, ArtStation quality, 8K in 2025 feels kinda crazy to me.
I’m not here to argue though. I just want to see the open-source community move forward.
But hey, do what you feel is right. 👍
@Jorot ahaha. yeah. i noticed on civit that some ppl still prompt like "masterpiece, best quality + tons of text in brackets like ((((big boobs:15.0))))"
I intend to continue promoting Chroma to the masses. First of all, I would like to thank Lodestone for the model, and I will try to continue doing what I do best. Ofc also big shout to all my fans, that supports me. Right now, my priority is Chroma and Qwen-Image, but if I have time, I might get around to wan and pony_v7
how do I use this lenovo with wan2.2? theres no low/high noise. do i just use this same lora in each lol
there's no high/low noise downloads, so do you just use the same safetensors in each low/high for wan2.2?>
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.