--Qwen
I have prepared a version for Qwen. It looks quite promising; however, work on its enhancement is still ongoing. Examples with girls i generated with my lora NiceGirls
P.S: yeah, I changed epoch in models manager, cause 7k steps gave overtrained result, and in process of testing I understood that 3k is enough
🗺️ Backstory
Lately I kept getting DMs like “When will you drop an iPhone‑style LoRA?” Problem: iPhone LoRA already lives on Civitai—and, well, I shoot on Samsung. SamsungCam UltraReal captures the S‑series look you can’t fake with generic “phone” prompts
After some tests i noticed that the best to use with UltraReal Fine-tune or UltraReal Abliterated with 60 steps (yeah, so much steps gives pretty good details, also text usually looks much better).
Use 0.7-1.0 weight (depends on artifacts in scene)
🎨 What it brings
Vivid, high‑contrast colors
Smartphone softness – just enough in‑camera smoothing to feel “real phone,” not plastic CGI.
Highlight‑happy HDR – clipped whites around street lamps for that over‑exposed‑yet‑cinematic pop.
Less bokeh, more context
🤝 NSFWloras-friendly
Tested with NippleDiffusion and MysticXXX (For some reasons is not visible in Suggested Resources)
You can try to generate in 2MP for more realism and details, but be ready that in some scenes hands will be broken. But still looks very good imo in usual 1MP.
In most scenes removes blurred background (sometimes it could be still blurred). Also improves details a lot even on default flux.dev (cause mostly i train all loras for my ultrareal fine-tune checkpoint
P.S: yeah, first version is slightly overtrained, but i glad with a result. v2.0 already WIP
1. 재게시된 모델의 권리는 원 제작자에게 있습니다.
2. 모델 원작자가 모델을 인증받으려면 공식 채널을 통해 SeaArt.AI 직원에게 문의하세요. 저희는 모든 창작자의 권리를 보호하기 위해 노력합니다. 인증하러 이동
