ArtificialGuyBR

Home / Blog / Training LoRAs

Fisheye Lens LoRA for Qwen-Image: Ultra-Wide Distortion Effects

Generate consistent fisheye lens distortion with this Qwen-Image LoRA. Trigger words, ComfyUI workflow, and example images.

1 sources cited Training LoRAs

On this page

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

Fisheye Lens LoRA for Qwen-Image

TL;DR

What This Model Does

This LoRA trains Qwen-Image to produce the distinctive fisheye lens look: extreme barrel distortion, curved straight lines, and an ultra-wide field of view that stretches edges toward the frame corners. It's a style LoRA, not a realism enhancer — the effect is artistic and consistent.

Training data consisted of fisheye photography. GPU compute was provided by Redmond.AI.

Trigger Words

Include in your prompt:

Fish Eye Lens. Fish eye camera.

The instance prompt used during training was exactly this phrase.

How to Use

ComfyUI (recommended):

  1. Load the LoRA via Load LoRA node
  2. Set LoRA strength (start at 0.8–1.0)
  3. Include trigger words in positive prompt
  4. Generate

Diffusers (Python):

from diffusers import DiffusionPipeline
import torch

pipe = DiffusionPipeline.from_pretrained(
    "Qwen/Qwen-Image-2512",
    torch_dtype=torch.bfloat16
)
pipe.load_lora_weights("artificialguybr/FISHEYE-REDMOND-QWENIMAGE")
pipe.to("cuda")

prompt = "Fish Eye Lens. Fish eye camera. A modern living room interior"
image = pipe(prompt).images[0]

Parameter | Value

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

LoRA weight | 0.8–1.0 Base model | Qwen/Qwen-Image-2512 Format | Safetensors License | Apache 2.0

Example Results

Images show consistent fisheye distortion across varied subjects — interiors, architecture, and outdoor scenes all exhibit the characteristic curved perspective. The effect is strong at full weight; reduce LoRA strength for subtler distortion.

Fisheye interior showing barrel distortion

Fisheye architectural exterior with curved lines

Common Errors and Fixes

Error | Cause | Fix

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

No fisheye effect | Missing trigger words | Add Fish Eye Lens. Fish eye camera. to prompt Over-distortion | LoRA weight too high | Reduce to 0.5–0.7 Artifacts at edges | Extreme distortion | Lower weight or adjust prompt composition

FAQ

Q: Does this work with other base models? A: Trained specifically on Qwen/Qwen-Image-2512. Results on other bases are untested.

Q: Can I use this commercially? A: Yes, Apache 2.0 license permits commercial use.

Q: What resolution works best? A: Qwen-Image native resolution (1024×1024 or 1024×1536). Fisheye distortion scales with resolution.

Sources