-- 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 (22)
amazing ! works well with liminal space images
Is there a trigger word?
yes. 8n8log
Amazing! When used with character loras, it makes them even better.
Beautiful! Did you train this one on bare Flux 1.D, or on one of the further trained models?
Thanx =) I trained on bare flux, but generated images with my ultrareal checkpoint
Is there a lora that removes this plastic look? Driving me insane. I guess I have to use Comfy to do it.
works great thanks
Works well at any strength, could be used to generally increase clarity at the 0.35 setting. Stronger and it starts to feel more like flux, but with the analogue feel the lora is good at. Only thing is in my tests at 1.0 it messed with the hands more, but it still kept strong coherence. I tested on chromav42-detailed at 30 steps with res_m+beta without NAG, the hyper loras, or caching. Prolly work still regardless of your different settings lol.
Cool 👍
As I understood, this author converted loras for chroma? I remember this author also published an Ultrareal LoRA for Chroma — though honestly, I didn’t test it, cause I love Chroma without LoRAs.
But maybe after the 50th version, I’ll go and retrain all LoRAs specifically for Chroma
@Danrisi I just thought it was you. At least you know about it. Honestly I only posted here because I thought you gave it to them
@makiaevelio543 I'm the author of the original LoRAs only — I didn’t try adapting them for Chroma myself. Honestly, I don’t mind if someone does — it’s promotion too, after all 😁
@makiaevelio543 lol, u know. just tested my loras at low power 0.35-0.5 wwith last Chroma and they are rly working good. thanx for tip
I have already tested Silver's conversion against the original LORA, and it barely makes a difference. I use the original one at 0.35 with Chroma (large), and it honestly changes the game. I won't run without Grainscape ;)
dreadfox it should simply prune the extra blocks the chroma model doesn't have
makiaevelio543 thanks!
This lora is a MASTERPIECE
Any plans of doing WAN 2.2 ITV/T2V? That film grain and Kodak look will blast the retro vibe looks :)
PS: great job! <3
Details
Files
GrainScape_UltraReal_v2.safetensors
Mirrors
GrainScape_UltraReal_v2.safetensors
GrainScape_UltraReal_v2.safetensors
GrainScape_UltraReal_v2.safetensors
danrisi_grainscape_ultrareal_v2.safetensors
camera-lens-kodak.safetensors
GrainScape_UltraReal_v2.safetensors
GrainScape_UltraReal_v2.safetensors
GrainScape UltraReal 2.0 for FLUX.safetensors
GrainScape_UltraReal_v2.safetensors












