This is a Midjourney Cinematic style Lora based on my aesthetics. Adds cinematic vibe and also additional detailing to a broad range of image themes without too much bias towards a single style.
Version 1: Trained on R128, 40 Images, 5000 Steps
FLOWMATCH NOISE FIX: ComfyUI beyond 0.3.77 changed the default UNET torch weight dType, for many users it used to default to fp32. FlowMatch is sensitive to this, and works best with fp32. To set the toggle, edit your .bat file for launching ComfyUI, and add to the end: --fp32-unet
Comparison article here: https://civarchive.com/articles/24424
Recommended workflow:
https://civarchive.com/models/2226355/luneva-infinite-details-z-image-workflow
Suggestion: Pair this with Luneva Cyber (HD Enhancer)
This other lora enhances ANY type of image into Ultra HD, improves composition, and also creates hyper realistic cyberpunk elements when prompted. Used in combination with Luneva Cinematic lora. <Download Luneva Cyber HD Enhancer>
Long Prompts:
If you're using very long prompts, you may need to increase the Lora strength (0.6 to 1.0) otherwise it locks onto your long prompt pretty strongly.
COMFYUI V0.4.0 BUG:
This version will break the EULERDISCRETE-SCHEDULER (makes your image very noisy again) However, this version is needed to run the Z Image Controlnet. So I've rolled back to v0.3.77 for now.
This is my very first Lora, hope you like it! Special thx to: @PastellPastellPastellI
Lora Strength Comparisons Below:
Description
v3.5: New version coming soon
FAQ
Comments (21)
真的很棒！
🔔 IF ANYONE HAS TROUBLE LOADING THE EULERDISCRETE-SCHEDULER reinstall/update the node.
rm'ing the eulerdiscrete node folder, and re cloning it fixed the issue for me. Thanks Luneva!
@blackfish95367 so it looks like it IS the node loading order. Since this error only occurs for some people. RES4LYF needs to fix this and not overwrite everyones schedulers.
❌❌❌COMFYUI V0.4.0 BUG:❌❌❌
This version will break the EULERDISCRETE-SCHEDULER (makes your image very noisy again) However, this version is needed to run the Z Image Controlnet, so I've rolled back to v0.3.77 for now and not using controlnet until it's fixed.
I'm sticking to 0.3.75 until the new UI settles down.
Thanks for your work, I love this Lora.
Works great with Forge Webui Neo!
bro this is absolute madness my guy
🏆🏆🏆 MY BOUNTY IS LIVE 🏆🏆🏆
Looking for BEST images for Cinematic Realism!
https://civitai.com/bounties/10855
LONG PROMPTS: If you're using very long prompts, you may need to increase LORA STRENGTH to 0.6-1.0 because long prompts lock the visuals a lot more.
HD ANIME LORA? I'm thinking about making a super high quality stylistic anime lora, particularly a semi-realistic hyper detailed style <example pic> that I was developing over a long period of time with Qwen. Currently it's impossible to replicate in Z Image despite my attempts, so it'll be massive amount of effort to figure it out and then generate perfect training data for it. Take weeks probably. I would only want to do it if there's really high interest in this, let's say, if this comment gets 30 Hearts.
FLOWMATCH NOISE FIX: ComfyUI beyond 0.3.77 changed the default UNET torch weight dType, for many users it used to default to fp32. FlowMatch is sensitive to this, and works best with fp32. To set the toggle, edit your .bat file for launching ComfyUI, and add to the end: --fp32-unet
Comparison article here: https://civitai.com/articles/24424
It has been so long and no new model has been released again.
me? I only want to release a something if it has significant positive impact to the community, I don't want to release something just to get some extra points. My next idea is for a really high end Anime lora, but I've been busy with some other things still =(
THANK YOU FOR SHARING THIS. please please pleease consider looking into making something for LTX2 - it so desperately needs something like this - and second, please consider the audio aspect as well, maybe something equally cinematic? <3
If I had a better GPU I would make a lot more stuff, it just takes me so long to do anything on my old GPU =(
@luneva i understand, same problem here, but i'ma workin on it! ..we all need to promote and push to buy decent cards for model creators! - now that would be a serious contest, say every 2 months, like a mid 40 or 50 series card or something!
This Lora and workflow is awesome for realism. Especially with the upscaler. My only question is, doesn't know any characters? Like American characters from TV and Movies or is it mostly just trained on modeling photos?
It wasn't specifically trained on any characters, but it would still have the original training done on the Z image Turbo model.
I love you ! I am so excited to try this !!!!
Boss, can you update the Lora based on the z image?
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.