ArtificialGuyBR

Home / Blog / Training LoRAs

Fisheye Lens LoRA for Z-Image-Turbo: Wide-Angle AI Images

Generate fisheye lens style images with Z-Image-Turbo using this LoRA. Trigger words, setup guide, and example outputs.

1 sources cited Training LoRAs

On this page

  1. What This Model Does
  2. How to Use
  3. Example Outputs
  4. Recommended Settings
  5. License
  6. Common Errors and Fixes
  7. FAQ
  8. Sources

Fisheye Lens LoRA for Z-Image-Turbo

What This Model Does

This LoRA (Low-Rank Adaptation) teaches Z-Image-Turbo to generate images with the distinctive fisheye lens aesthetic — ultra-wide field of view, strong barrel distortion that curves straight lines toward the edges, and a characteristic central magnification effect. The training data consists of fisheye photography, so the model reproduces the optical signature of actual fisheye lenses.

How to Use

  1. Load the LoRA in ComfyUI (recommended) or your preferred diffusion interface
  2. Include the trigger phrase: Fish Eye Lens. Fish eye camera.
  3. Adjust LoRA weight — start around 0.8–1.0 and tune for your subject
  4. Generate using Z-Image-Turbo as the base checkpoint

The model outputs Safetensors format weights compatible with diffusers and ComfyUI.

Example Outputs

Fisheye lens style generation showing curved architectural lines and wide field of view

Fisheye distortion effect on interior space with characteristic barrel distortion

Outdoor fisheye perspective with curved horizon and expanded field of view

License

Apache License 2.0 — permits commercial use, modification, and distribution with attribution.

Common Errors and Fixes

Error | Cause | Fix | Source

|-------|-------|-----|--------

No fisheye effect | Missing trigger words | Add Fish Eye Lens. Fish eye camera. to prompt | HF model card Over-distorted results | LoRA weight too high | Reduce weight to 0.6–0.8 | Community testing Base model mismatch | Wrong checkpoint | Use Tongyi-MAI/Z-Image-Turbo | HF model card

FAQ

Q: Does this work with SDXL or Flux? A: No, this LoRA is trained specifically for Z-Image-Turbo architecture.

Q: Can I use it commercially? A: Yes, Apache 2.0 license allows commercial use with attribution.

Q: What's the recommended LoRA weight? A: Start at 1.0 and reduce if distortion is excessive; 0.8 is a good baseline.

Sources