-- Z-image
I generate with res2s_ode/res2m_ode + beta. Using with strength 0.85-0.9 and using guidance 3.5. Feel free to take my workflow from my images
V2.0
This updated version was trained on a completely new dataset, built from scratch to push both fidelity and personality further.
Vertical banding on flat textures has been noticeably reduced—while not completely gone, it's now much rarer and less distracting. I also enhanced the grain structure and boosted color depth to make the output feel more vivid and alive. Don’t worry though—black-and-white generations still hold up beautifully and retain that moody, raw aesthetic.
Also fixed issue with the same face.
Think of it as the same core style—just with a better eye for light, texture, and character.
P.S: this time trained in usual 1024, not 2048
Forget soulless, digital perfection - this LoRA drags your Flux generations straight through a box of expired Kodak Tri-X and leaves them beautifully battered. Grainy, noisy, and full of life, this model captures the raw, textured magic of analog film photography. Think vintage press photos, forgotten negatives in a thrift store, or that one roll of film you shot in college and never developed—until now.
Why This LoRA?
📸 Trained at full 2048×2048 resolution.
🎞 Authentic film grain – Not some fake digital overlay. This is deep, baked-in texture.
🖤 Black & white mode goes hard – Deep shadows, rich highlights, and that classic documentary grit.
🌿 Cinematic depth of field – Backgrounds are blurred naturally, like real fast-aperture film photos. But unlike digital bokeh, it’s not surgically perfect - it has that slight unpredictability, just like an actual 35mm lens at f/1.8.
P.S: Honestly? I’m not a huge fan of shooting on real film - especially when there aren’t many places in my city to develop it. But this LoRA isn’t about the hassle of film; it’s about the look I actually like. My own vision of analog photography - without the wait times, processing costs, or tragic “accidentally opened the camera back” moments.
If you want AI generations that feel like they were ripped from a forgotten archive of classic film photography, GrainScape UltraReal is your go-to. Perfect for photographers who miss the days of pushing film to 1600 ISO - or anyone who just wants that timeless, nostalgic imperfection in their AI art
Description
FAQ
Comments (16)
It's crazy good. I just hope that you introduce these cinematic realistic styles in your ultrareal finetunes also
Hey, I just found out the new wan 2.1 model can also generate images. And it's MUCH BETTER as it is not distilled. I think if you finetune it,the improvements would be great. Also, have you tried using flux dev dedestilled or flex lite?
Really? Could you link to where you saw the wan 2.1 images? Here's on civit there's only a lot of porn shorts lmao.
And, by Flex lite, you mean this?: https://civitai.com/models/1246082?modelVersionId=1404610
If so, Danrisi mentioned looking into fine-tuning that in their UltraReal v4 post.
@Lathos https://www.reddit.com/r/StableDiffusion/s/K0uzYwTxxb
@aleemawan118373 hmm, sounds interesting. Need to learn a little bit how to finetune wan
@aleemawan118373 Those are CRAZY good! I didn't expect a video model to be able to produce Flux-level images, even surpassing it in some cases considering how natural these shots look. Do you think it supports controlnets such as tile, depth, canny, etc? If it does, it has a great potential to rival Flux in popularity.
@mmdd2543 The github diffusers team is working on implementing wan into their library. They have 2 pipelines for T2V and I2V. So I would presume you may (in the future) be able to utilize the flux controlnets as well as additional samplers.
I was about to upload this lora on Seaart but found out you have a profile there... Could you upload this lora on Seaart please? :)
I've been getting an error about your workflow file for a few days, already installed comfyui again but nothing is changed. do you have any idea?
https://imgur.com/a/j1ee4N0
Weird error, didn't see such before. What version of comfyUI u have?
@Danrisi v0.3.29
@Danrisi seems the problem is "anything everywhere" node. its not working as it should be.
@kurtdoe strange, i have 0.3.29 too. yeah, maybe problem is wwith this node, cause i remember long time ago already had some issues with this node. but now everything work good for me. but u can try to reconnect manually vae, clip and model with loras
@Danrisi yes i just wrote you few hours ago. its the "anything everywhere" node. your workflow works solid. thanks for the response.
@kurtdoe sorry, this month i have very little amount of time i can spend on civit and trainings
@Danrisi no worries, everything works fine. cheers.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.
















