Z-Image_REAL_Turbo_SaMay_bf16

This is a Z-Image Turbo model merged with my Realistic_LoRA.

Please place it in the Models/Unet folder.

Realistic_LoRA was trained on 5,000 real photos with a focus on natural lighting, realistic skin and hair, and small details like hands and background depth.
The goal is simple: push the base model closer to clean photographic realism without turning everything into plastic beauty shots.

Training & Tech

• Dataset: ~5k real-world photos (people, objects, everyday environments)

• Target: general realism, not tied to a single art style

• Base: merged with Z-Image-Turbo bf16 for fast inference

• Use cases: portraits, half-body and full-body shots, lifestyle, simple fashion, product photos

You can stack this LoRA with your own style LoRAs or strong prompts – it’s designed to enhance realism, not overwrite the look of your pipeline.

Credits & NexTry

Training was funded by NexTry.app (VTON) – a virtual try-on startup.
If you want to see how realism holds up in a real product, you can:

• try changing clothes on your own photos at NexTry.app,

• see how it handles lighting, body shapes and different fabrics,

• and leave honest feedback – it helps us train the next open models.

Collaboration

I’m open to custom training / fine-tuning for:

• brand- or product-specific realism LoRAs,

• internal tools, demos, or research projects,

• experimental models and content pipelines.

If you test this model, please share example images and your settings – sampler, CFG, steps, etc. It really helps to understand how it behaves in different setups.

---

License

This model is released under the Apache 2.0 License.
Based on Tongyi-MAI/Z-Image-Turbo (Apache 2.0) by Tongyi-MAI.