Description
This page includes all currently available versions of my Q7 model series.
If this is your first time seeing this model, you can check the preview images and choose the version that best matches your taste.
About the Model Series
Sorry for the long pause between updates. Recently I have been running a bit low on inspiration, so I have not had many new works that felt good enough to share.
This series has gone through multiple rounds of merging and adjustment. Different versions aim at different balances between anime style, 3D rendering feel, realism, lighting, color distribution, facial effects, and overall compatibility with LoRAs.
Some versions are more experimental, while others are meant to be more stable and practical for daily use.
Latest Update
LoraModel-v3.1
The main purpose of this model is to provide better compatibility for the Q7 series of LoRAs. It can accurately reproduce both realistic and anime styles. I have made appropriate adjustments to ensure it generates good images even when no LoRA is loaded. I haven't tested its compatibility with other LoRAs yet. Its base derived model is llustrious-v2.
q7-remix-triple-v1
2026-03-26
This Triple V1 model is built through a new merging approach. While maintaining strong overall visual quality, it also offers better compatibility with other LoRAs than my previous models. In my opinion, it performs better than the earlier ones in both balance and usability.
The overall style is roughly 70% anime and 30% 3D. If you like it, feel free to show me your images. I would love to see them and use that feedback to guide my future work.
Recommended CFG: 4
Also, some of my other models do not have a VAE merged in, which may result in gray-looking images. I have posted a specific solution for that in the comments section.
Version Guide
v1
Version 1 has a strong cartoon style, but the overall image appears somewhat hazy.
v1.3
A minor tweak based on version 1.0, with filter-related effects removed.
v1.5.1
To be added.
v2.0
Based on Version 1, Version 2 reduces the proportion of the cartoon-style filter, adds more anime-style influence while keeping the overall tone unchanged, and features a richer color distribution.
v2.1
Based on Version 2, Version 2.1 adjusts the character body shape, skin texture, and lighting. It leans more toward a realistic style, with the cartoon feeling significantly weakened.
v2.2
Based on Version 2.1, the main adjustments are focused on visual effects. The advantage of Version 2.2 is that it can output images with a near-realistic style while still preserving anime-like composition.
Highlight: Highly compatible with various character-type LoRAs.
v2.4-unstable
This version incorporates some previous LoRA styles to achieve better environment, lighting, facial effects, and body effects. Of course, it may still contain some bugs, but I had not found any obvious ones at the time, so it was originally marked as unstable.
v2.4-stable
Version 2.4-stable fixes the issues found in the unstable version and was re-uploaded as the refined release. However, the unstable version still has some unique qualities, so I decided to keep both versions available.
TripleV1
A tentative step, primarily 3DCG/anime style.
LoraRemix
add a few lora with 0.3
LoraModel-v3.1
designed for better compatibility with the Q7 series of LoRAs
Notes on Use
About compatibility with LoRAs
Some versions in this series are designed to maintain decent compatibility with other LoRAs. However, actual performance still depends on the LoRA itself, prompt structure, and sampling settings.
If you are stacking multiple LoRAs together, I recommend keeping the total weight under control rather than forcing every LoRA to run at full strength. In most cases, lower and more balanced weights will give more stable results.
About VAE
Some of my older models do not have a VAE merged in. If you encounter gray-looking or washed-out images, please check the comments section where I posted a specific solution.
Recommended Settings
Recommended CFG for q7-remix-triple-v1: 4
For other versions, you may need to adjust settings based on the style you want
Personal Note
Please do not pair generated images with nonsensical sexually suggestive text and post them on Pixiv just to bait or mislead people into following you.
To me, the value of AI image generation lies in openness and sharing. Even a so-called “bad image” can still have something interesting in it. That has always been the attitude I try to keep.
It is not that attention means nothing to me, but creativity and inspiration are far more valuable. So if you create something with this model, feel free to share it.
And if you ever have new ideas, feel free to leave a comment.
Description
This version underwent additional training to fix the texture detail blurring issue in version V2.
FAQ
Comments (6)
2.2 just doesnt work for me, generates fine, but last frame makes it appear grey with strange white bloom dots. Guess im missing vae
Regarding the VAE issue, I didn't embed VAE into the model. This is a personal preference of mine because some VAEs can alter the colors of the images generated by the model.
This problem is easily solved. Depending on your platform, if you're using ComfyUI, you'll need to add an extra node that reads the VAE and connects to the K-sampler. If you're using a web UI, you can read it directly.
This is a generic SDXL model, VAE.
@q7777 Can you tell me what workflow you are using?
@neovnchoi700 You mean T2I? I don't use ComfyUI's workflow. It's great and highly customizable, but it's not crucial for me, so I'm still using WebUI. I recommend trying this: https://space.bilibili.com/12566101?spm_id_from=333.337.search-card.all.click
请问什么时候更新2.3呢
2.3遇到点问题，它本应该是很多种不同的面部特征但是被我融合成一个了，还在尝试修复这个问题，但是每个版本都不和我心意，大概还需要一阵子。