PLEASE READ THE VERSION NOTES
What this is?
A turbo checkpoint merged for the perfect balance between three things: NSFW capability, prompt adherence, and a "smartphone-like" look. The goal is a balanced crisp, casual image. This is the goal, and this checkpoint is the starting point; expect many updates with high frequency.
About FinePorn
This is an approachable but functional starting point. To hit the target, I baked in LoRAs with balanced weights + INT8 and CONVROT to achieve faster speed and good quality, with the plan to scale up future versions by adding new LoRAs, rebalancing weights and hunting for the optimal settings.
Baked LoRAs Changelog (v4):
Realism Engine v3 — balanced
Lenovo v1 — removed due to dark contrast
NSFW V4.3 — balanced
PhonePhoto 2025 — removed due to a variation issue
Elusarca's Krea 2 — added
UltraReal Breasts&Nipples — added at very low strenght
Aesthetic Amateur Photo — added at very low strenght
Breast Squeeze — added at low strenght to enforce interactions
Wetness Slider — added at negative strenght to reduce dropplets
All baked LoRAs will credit their creators.
⚙️ Recommended Settings
V1
res_2s + beta -> More Quality
er_sde + beta -> More Speed
CFG: 1
Steps: 8–10
V2
SamplerER_SDE + beta -> Fast and Stable
CFG: 1
Steps: 8–12
V3
euler + simple
euler + beta (my favorite)
er_sde + simple
er_sde + beta
V4
euler + beta
er_sde + simple
📐 Resolutions
I found that scaling resolutions by x1.25 and x1.5 gives better results:
- 832x1216 standard → 1040x1520 optimal 1248x1824 recommended
- 896x1152 standard → 1120x1440 optimal 1344x1728 recommended
- 1024x1024 standard → 1280x1280 optimal 1536x1536 recommended
💡 Notes
Good results are achievable without a face detailer or extra passes. With the right config, some patience, and a bit of seed luck, you can get great things.
I highly recommend start all your prompts using a phrase like this or similar.
This is a casual, low-quality photothis is an amateur photo taken from smartphone, casual photo⚠️ Important disclaimer
This is NOT trained by me and is NOT my own work, my job here is using my experience to bake balanced weights of my personal taste to achieve desired images using my prompt style. My only role is to share with you what, to my eye, is the best style and quality achievable with this model. All baked LoRAs credit their creators.
ABOUT PROMPTS
I'm a bit dissapointed tbh, i've seen a dozen of 'big' creators and model makers just copying my hand-made prompts and spiting them on their showcase photos, with zero quotes to me or at least a small comment or a dm, it feels bad and very lazy coming from them.
so, from now on, complex prompts will be shared only on DMs freely, but i'll keep simpler prompts public.
Description
About this version
Rebalanced LoRas (v3):
Realism Engine v3 — updated+reduced weight
Realistic Snapshot — reduced weight
NSFW V4.3 — updated+increased weight
SNOFS V1.1 — removed
RawGirlSpicy V2 — removed
PureLens V1 — removed
PhonePhoto 2025 — updated+balanced
UltraReal Breasts&Nipples — updated+balanced
EXPLANATION:
I think I went overboard with the excessive LoRas in V2, causing the loss of Asian features and an excess of surgically altered faces and unwanted textures. I also feel that the homemade photography style was somewhat lost in V2. Therefore, in V3 I'm aiming for a version closer to V1, but with updated and better-balanced LoRas to achieve better adherence, a high level of explicitness, fidelity to features and ethnicities, and a more homemade and less artificial aesthetic.
i encourage you to experiment with different forms of prompts to achieve the same result).
NEWER IS NOT BETTER
I want to clarify something, this is my subjetive taste, so V3 isn't necessarily better than V1/2, there could be the case where you prefer to keep on V1/2 because it fits perfect for you.
ABOUT WORKFLOW
For all these images i simply used Native KSampler with the settings Euler/Er_sde + Simple/Beta. I still experimenting a new V3 workflow, but i'm not convinced to release it yet because feels very experimental so i prefer to maintain the things as simple as possible.
ABOUT PROMPTS
I'm a bit dissapointed tbh, i've seen a dozen of 'big' creators and model makers just copying my hand-made prompts and spiting them on their showcase photos, with zero quotes to me or at least a small comment or a dm, it feels bad and very lazy coming from them.
so, from now on, complex prompts will be shared only on DMs freely, but i'll keep simpler prompts public.
Nothing more to say, i hope this can be useful for you and if you like it, please leave a comment, it's useful for me :) enjoy.
Sincerely, Simon.
FAQ
Comments (37)
When generating latina / darker skin characters, it always generates them with tan lines. I was not able to fix this with prompting
give me an example prompt and i'll try to solve it
Niceeeee, downloading right now, sample images look very good.
thanks, i hope it fits well for you
Truly outstanding work.
i'm glad you liked it
niceee, i will try
I think Euler Simple/Beta is giving better results for V3, just my subjetive guessing
V3 obsessively adds water marks and water droplets to the models' bodies.
hmmm i'm not having this problem, what settings are you using?
@simonishere "er_sde,beta,10 steps." Now I have to use the "humidity slider" to control this.
@simonishere I rebuilt the workflow, and the issue was resolved. Although I never figured out what the problem was.
Same issue here, no clue what it is.
@alex9441 @ficohito2343 i think er sde tends to do this often, euler should be better
@simonishere I'm having the same issue as my peers, and switching to Euler simple/beta doesn't fix it. in fact, the models end up with warts! The first two versions worked perfectly with my workflow, but this one seems a bit more finicky. ^
@simonishere I think it's because of Realistic Snapshot. When I use it with an other checkpoint, I have the same issue with droplets.
@KinkReaper interesting, i gave that lora the lowest weight but could be, i'll see if i can come with some fix
@simonishere Thanks. Your work is incredible by the way!
@KinkReaper uploaded a fixed version without snapshot lora and added lenovo, feels better
@simonishere It's just perfect now! Nice we have found the culprit! Thanks!
Any chance we got an FP8 of the V3?
already uploaded fp8 on the new v3 fixed
Fixed seems to be working better, more prompt adherence and better image quality, less unwanted stuffs and also the fp8 is available.
hoping your feedback
Interesting that a prompt like "low quality" helps get rid of that "AI look" a little bit.. I'm giving this model a try.
yes hahaha, that's because the captioning of some loras uses low quality for smarphone-like photos
@simonishere oh, makes sense then.. but yeah, that's a nice prompt to have to combat the "AI perfection" ;)
@ArtiOtaku that's exactly what i want to achieve with this model haha
When I load the int8 version, the filename is the same as the fp8 version.
I don't know if that's a mistake on Civitai's end, but I just wanted to let you know in case you can change it.
By the way, great model!
the int8 has a '_INT8' added on the end, but maybe civit is breaking it, i see it well tbh, but thank your for pointing it and thanks for your comment! :)
Best checkpoint!
Thank you!!
Hello! Interesting model. In many cases, you don't even need to use a censorship-removal LoRA since the model handles it well on its own. Thanks for sharing!
Thank you bud, that's my main target, a model wich can handle several nsfw concepts without needing of loras
Hey ma dude great work, could you by any chance share the weights (intensity) of the LoRAs you merged, privately? If not, no worries!
NVM found it in metadata!
V3.1 has better color, but its skin detail falls short of V2; V2 remains the best in terms of realism.
No idea why, V3 doesn't follow my prompts at all, but V2 is awesome, my go to. Thank you