ArtificialGuyBR

Home / Blog / Training LoRAs

Polaroid Redmond Z-Image LoRA: Instant Film Style for Z-Image-Turbo

LoRA that adds authentic Polaroid instant-film aesthetics to Z-Image-Turbo. Trigger words, usage guide, and example outputs.

1 sources cited Training LoRAs

On this page

  1. What this model does
  2. Trigger words
  3. How to use
  4. Recommended settings
  5. License
  6. Example outputs
  7. Common errors and fixes
  8. FAQ
  9. Sources

Polaroid Redmond Z-Image LoRA

TL;DR

What this model does

This LoRA (Low-Rank Adaptation) modifies the Tongyi-MAI/Z-Image-Turbo base model to generate images with the distinctive look of Polaroid instant photographs. The style includes the classic white border, characteristic color palette with slight warm/cold shifts, subtle vignetting, and the organic imperfections that make analog instant film recognizable.

The model was trained on a curated dataset of Polaroid-style images using GPU compute sponsored by Redmond.AI. It maintains the strong prompt adherence and speed of Z-Image-Turbo while layering on the instant-film aesthetic.

Trigger words

Include both phrases in your prompt:

Polaroid Photo, PolaroidRedAF

These tokens were used during training as the instance prompt and reliably activate the style.

How to use

  1. Download the .safetensors file from the Hugging Face repository
  2. Place it in your ComfyUI/models/loras/ folder
  3. Add a Load LoRA node to your workflow
  4. Set the LoRA strength (start at 0.8–1.0)
  5. Include the trigger words in your positive prompt

diffusers (Python)

from diffusers import DiffusionPipeline
import torch

pipe = DiffusionPipeline.from_pretrained(
    "Tongyi-MAI/Z-Image-Turbo",
    torch_dtype=torch.float16
).to("cuda")

pipe.load_lora_weights("artificialguybr/POLAROID-REDMOND-ZIMAGE")

prompt = "Polaroid Photo, PolaroidRedAF, a sunset over mountains, highly detailed"
image = pipe(prompt).images[0]

Parameter | Value

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

Base model | Tongyi-MAI/Z-Image-Turbo LoRA weight | 0.8–1.0 Sampler | Euler a / DPM++ 2M Karras Steps | 4–8 (Z-Image-Turbo is a turbo model) CFG | 1.0–2.0 Resolution | 1024×1024 or 512×512

License

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

Example outputs

Polaroid-style landscape with white border and vintage color rendering

Portrait in Polaroid aesthetic showing characteristic color shifts

Common errors and fixes

Error | Cause | Fix | Source

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

Style not appearing | Missing trigger words | Include "Polaroid Photo, PolaroidRedAF" in prompt | HF model card Overcooked look | LoRA weight too high | Reduce to 0.6–0.8 | Community feedback Wrong base model | Using SDXL or SD 1.5 | Must use Tongyi-MAI/Z-Image-Turbo | HF model card

FAQ

Q: Does this work with SDXL or Stable Diffusion 1.5? A: No. This LoRA is trained specifically for Z-Image-Turbo and will not produce correct results on other base models.

Q: What's the difference between this and other Polaroid LoRAs? A: This targets Z-Image-Turbo (a fast turbo model), not SDXL or SD 1.5. The trigger words and training dataset are specific to this adaptation.

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

Sources