# 🔥 ONE MEMBERSHIP = FULL ACCESS
Your membership includes :
- models
- workflows
- scripts
- Exclusives
- future updates
👉 $25/month on Ko-Fi
📢 Join The Community
We're building a friendly space to chat, share creations, and get support.
👉 Click here to join the Discord
✨ IntoRealism Ultra Amateur is an experimental release that combines the strength of DMD2 with the finesse of a Light model, delivering a more realistic look than ever before.
👉 A rare feature : unlike most light models, negative prompts work perfectly, giving you much more control and precision over the final renders.
🔹 Key features:
DMD2 + Light blend
Enhanced realism (textures, skin, natural lighting)
Handles NSFW content very well
Fully compatible with negative prompts
Versatile: portraits, lifestyle, cinematic scenes
better LoRA adherence
⚠️ Note : this is an early test version. The model will be further optimized and refined in future updates for even greater realism and flexibility.
Usage
Sampling method : Euler A Karras or Exponential
Sampling steps : 6 - 12
Resolution : 896x1152 / 832x1216
CFG : 1.5 - 2
Description
FAQ
Comments (21)
I like how your checkpoints/models are evolving... nice work! This one seems nicely "brighter" with good lighting. I'll post an image as well. Hat tip to @GBRX for the cool image prompt ideas!
just started using this model and i'm absolutely loving it, can't wait to see the full release
edit: after extensive testing my only complaint is that this model only generates ugly circumcised penises, please add more variety to penises
Prompt adherence is SOLID.
Interesting. Lcm is too smooth, Eulera is too grainy.
Very realistic with euler though. I wont say prompt adherence is good(huge asses only, when i prompt small round butt).
small butt + slim physique in your prompt might help, i've had really good results with it
Too bad, good foundation but major problems with fingers and feet.
are there any specifics training loras on DMD2 checkpoints? I tried once, it was washed out and stuff, seems like a great base NSFW to try&train on
Training my LoRA's on SDXL has worked perfectly with DMD2.
@realdunderton no matter if normal or DMD2 checkpoints?
@NeuroFunkeR I haven't had any issues with them on either SDXL or DMD2 checkpoints.
@realdunderton k thx ill try then
@realdunderton ì have a functioning SDXL lor trining preset, as soon as i start trainin on DMD2 it gives me washed out results. With and without DMD2 lora even.
i tried training a character lora on this and it came out pretty bad
batch size:4
epochs:100
steps:10k
images:200
@NeuroFunkeR Weird. I trained on collab using 1024x1024 dataset with SDXL 1.0 as the training model; it has worked well for me with DMD2 models.
I would give more lighting for Euler or more dmd for Lcm. Its not working good with Euler in present state.
Euler A, not Euler.
@enzino Yes, of course. Its too noisy. I like your model, thatswhy i wrote this. Your previous has the same problem for me. Somebody likes this noisy pics, but you can just add with prompt or lora to it.
I think you may have re-uploaded the Realism Ultra model by mistake. I already had that one downloaded and in my checkpoints directory, and when I downloaded the model attached to this page and dropped it in my folder, it asked if I wanted to replace the one I already had.
Nop, it's the right file. It was renamed for some reason I don't know on Civitai. Feel free to add “Amateur” to the name to differentiate it.
@enzino Got it! Thanks for the quick response
Euler a, CFG 1, 8 steps is my preference for detail. adding "texture" to the beginning and then "top down close-up" at the end makes insanely detailed results of a realistic pic. Unlike most lightning DMD2 models, negative prompts are very responsive, so I am sure many flaws can be fixed using that to your advantage.
I have over 30 SDXL models in my library and this is by far my favorite SDXL model over all of the others. I feel like the tags could use a bit of steering (e.g. "creampie", or "cum inside" is a messy drip effect but "semen inside" makes your expected result) but using slangs would be extensive for training, I could imagine. I am really excited to see where the future sits with this as I could see it being an insane foundation for hyperrealism and amateur photos in general.
The other IntoRealism Ultra model is pretty tampered when it comes to a grainy amateur effect, so this was perfect. For a v1 model this was pretty insane. Great job, as always ;)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.