TJFilmStyle is a style LoRA I specifically trained to simulate real film photography, featuring the unique texture of physical film. It enhances the lighting, texture and close-up depth-of-field effects for landscapes, still lifes and portraits alike. Especially for portraits, it can significantly improve skin texture, replicate the translucent skin look of film photography, reduce the AI-generated artificial feel, and deliver more outstanding light and shadow performance.
Nov.6 2026 updated MossImage V2:
V2 improves the training method, optimizes the training set, and enhances the training accuracy based on V1, in order to improve the graininess of simulated real film.
Compared to V1, V2 maintains excellent lighting and high versatility, resulting in a more stable, realistic, and noticeable graininess of the film, but with slightly different color tones.
Recommended sampling method: Euler+Simple, with 25 steps, CFG=3
Welcome everyone to test and return more pictures.
V2是在V1的基础上，改进了训练方法，优化了训练集，提升了训练精度，用以提升模拟真实胶片的颗粒感。
相比V1，V2保持优秀光影、高泛用性的条件下，胶片颗粒感更稳定更真实更明显，但色调略有不同。
推荐采样方法：Euler+Simple，步数：25步，CFG=3
欢迎大家多测试、多返图。
Oct.18 2026 updated MossImage :
This Lora is my TJfilm style Moss version, trained on Moss Image Base, suitable for both Moss SFT and Base versions of text2images. Compared to Krea2 and Zimage, I feel that the TJfilm style in Moss Image has more vibrant colors and also good stylization ability. My test image adopts 30 steps, cfg=3.0, generated by Euler+simple, which makes it convenient for me to quickly generate the chart.
I suggest everyone try the official recommended steps and cfg, and I also welcome everyone to return more pictures.
Of course, this is a fast Lora training version of my MossImage, but there are still some shortcomings.
这个lora是我的复古胶片风格的moss版本，在moss image base上训练，适用于moss sft和base版本的文生图。相较于krea2和zimage，moss image下的tjfilm风格我感觉颜色更艳丽，风格化能力也不错。我的测试图采用30步，cfg=3.0，Euler+simple生成，方便我快速出图。
我建议大家可以试试官方推荐的步数和cfg，同时我也欢迎大家多多返图。
当然这是我mossimage的一个快速lora训练版本，仍有不足。
Aug.22 2026 updated Krea2 V2.0 Features:
The dataset has been optimized, and training parameters have been re-adjusted.
Delivers more stable and superior performance in lighting and atmosphere rendering.
The film grain texture presentation has been significantly improved.
V2.0特点：
1.优化了训练集，重新调整了训练参数。
2.更稳定更好的光影氛围表现。
3.胶片颗粒质感表现更好。
Compared with ZImage, Krea2's version demonstrates better prompt adherence, enhanced stylization capabilities, and superior aesthetic performance.
Recommended Settings:
Trigger word: vifilm
weight:1
Sampler: Euler, Scheduler: Simple/ddim_uniform
The Krea2 version can already render grain texture excellently. If you want to enhance the grain effect, you can add the prompt word "film grain".
Prompt Tips: When generating images, try to avoid piling up and overlapping style-related prompt words, so as not to undermine the generation quality. It is recommended to use natural language descriptions instead.
TJFilmStyle是我专门训练，用来模拟真实胶片摄影的风格lora，带有胶片特有的质感，无论是风景、静物、还是人像，光影、质感以及近景的景深效果都得到了增强。尤其在人像方面，能明显提升皮肤质感，模拟胶片摄影的皮肤通透感，减少ai感，光影更加出色。
相比zimage，krea2的lora提示词遵从度更好，风格化能力提升，美学表现更加优秀。
推荐参数设置：
触发词: vifilm
推荐权重：1
采样器Euler 调度器Simple/ddim_uniform
krea2版本已经能很好表现颗粒感，如果想加强颗粒感，可加上提示词film grain。
提示词使用注意：生图时尽量避免风格类提示词的堆砌、叠加使用，以免影响生成效果，推荐采用自然语言描述。
Description
V2 improves the training method, optimizes the training set, and enhances the training accuracy based on V1, in order to improve the graininess of simulated real film.
Compared to V1, V2 maintains excellent lighting and high versatility, resulting in a more stable, realistic, and noticeable graininess of the film, but with slightly different color tones.
Recommended sampling method: Euler+Simple, with 25 steps, CFG=3
Welcome everyone to test and return more pictures.
V2是在V1的基础上，改进了训练方法，优化了训练集，提升了训练精度，用以提升模拟真实胶片的颗粒感。
相比V1，V2保持优秀光影、高泛用性的条件下，胶片颗粒感更稳定更真实更明显，但色调略有不同。
推荐采样方法：Euler+Simple，步数：25步，CFG=3
欢迎大家多测试、多返图。