This is a fully SFW checkpoint! No NSFW content！
It features significant aesthetic upgrades with a unique artistic style of its own, and it works universally across all art styles and functional generation scenarios.
It feels like I’m the only one in the community who still hasn’t given up on Z Image Base.
这是一个SFW检查点！仅有原版Base级别的NSFW内容！
美学方面提升很大，同时又有自己的风格。能影响到所有的风格和功能。
感觉现在整个社区只有我还在没有放弃Z image base。
Recommended Params
Steps: 8–20，CFG Scale: 1–2，Samplers: heunpp2 / Euler a / Seeds_3 / DDIM，Schedulers: beat / linear_quadratic / ddim_uniform，Upscaler: 4x-UltraSharp
Personal Optimal Settings：10 steps, 1.5 CFG, heunpp2 + beat, with 4x-UltraSharp upscaler
This ZIB checkpoint has been my daily driver for the past two months. I simply adore ZIB outstanding diversity and prompt adherence — strengths that ZIT simply does not possess.
This checkpoint is integrated with speed-optimized LoRA. While generation speed is greatly improved, it comes at the cost of reduced general LoRA compatibility. Using external LoRAs will lead to degraded image quality and unsatisfactory results.
这是我这两个月一直在用的ZIB检查点，我就是单纯的喜欢ZIB的多样性和提示词准确度这些ZIT没有的优点。
这个检查点因为合并了加速lora，速度提升的同时也对lora的兼容性有影响。使用其他lora只会得到更差的结果。
If you find this checkpoint useful and wish to support my work, or if you only need the standalone LoRA version, please consider donating or purchasing the LoRA on TensorArt.
All donations will be reinvested into developing higher-quality checkpoints for everyone. I’m also eagerly looking forward to the open-source release of the Z Image Edit version.
所以，如果觉得好用，或者只想使用lora版本请到Tensorart打赏我或购买lora。我会利用这份捐助给大家带来更好的检查点的，我现在非常期待Z image edit版本开源。
Description
Recommended Params
Steps: 8-20 | CFG Scale: 1-2 | Samplers: heunpp2 / Euler a / Seeds_3 / ddim | Schedulers: beat / linear_quadratic / ddim_uniform | Upscaler: 4x-UtraSharp
My go-to settings for it: 10 steps, 1.5 CFG, heunpp2 + beat + Upscaler: 4x-UtraSharp
FAQ
Comments (8)
The model is really great. People misunderstand that ZIB is worse than ZIT. In my opinion, ZIB is a model that is sensitive to users' skills. Thank you. I also bought LORA
Thank you for your support. bro 💛
Did you also merge "Z-Image Base-Art Scene" into this checkpoint?
Yes, I first merged my four LORA images using z-image merger nodes, then fused them with checkpoint, finally, merge accelerate Lora.
@Ray3780 I've been using it all day today, and it's really great. Especially, it's great at the same time, it's aesthetic. I'll continue to support you in the future
ZIB is just underdeveloped, so I hope it will spread
I think ZIB has consistently outperformed ZIT, but training LoRa has bugs and overly stringent conditions, resulting in generally lower LoRa quality, which is why it's disliked.
My ZIB is worth a try. I invested a lot in training it, almost always with over 100k steps and a training set of 1k-3k images.
To date this is the only ZiB checkpoint I tried that produces good results on its own, without the need for a ZiT refinement. Great job!