Z-Image-Turbo-PNTE-Negative-LoRA
Model description
Overview
This LoRA is designed to be applied at negative strength values with diffusion models. When used correctly, it increases perceived detail density and overall image quality.
This variant is trained specifically for Z-Image-Turbo.
Intended Use
Manually set the LoRA strength to a negative value (e.g. `-0.1` to `-1.0`).
Start at a low value and gradually increase the strength to achieve the desired level of detail while using a locked seed for comparison.
Applying positive strength is not the intended use case. Positive values tend to simplify results and reduce detail.
Training Data
This LoRA was trained on the seed images used in the COCO CLIP R-Precision evaluation set associated with OpenAI’s Point-E (2022).
The dataset consists of evaluation seed images. The training objective was not aesthetic stylization, but behavioral modification under negative application.
Disclaimers
- No affiliation with OpenAI.
- Results may vary depending on sampler, CFG scale, resolution, and prompt structure.
- More examples coming soon.
Download model
Download them in the Files & versions tab.