ArtificialGuyBR

Home / Blog / Training LoRAs

App Icons LoRA for Z-Image Turbo

Generate consistent app icons with this LoRA trained for Tongyi-MAI Z-Image Turbo. Learn trigger words, settings, and setup.

2 sources cited Training LoRAs

On this page

  1. What This Model Does
  2. Trigger Words
  3. Recommended Settings
  4. Setup
  5. Credits
  6. Common Issues
  7. FAQ
  8. Sources

App Icons LoRA for Z-Image Turbo

Generate consistent, clean app icons using a LoRA trained for Tongyi-MAI's Z-Image Turbo.

What This Model Does

ICONS REDMOND for Z-Image Turbo is a LoRA adapter trained on a curated dataset of mobile app icon designs. It adapts the Tongyi-MAI Z-Image Turbo base model to produce square, flat-design icons with bold colors and crisp vector-like aesthetics — the kind you see on the App Store or Google Play.

The training focused on icon composition, color consistency, and the clean simplicity that makes app icons recognizable at small sizes. Outputs are 1024x1024 pixels, matching standard icon dimensions for high-DPI displays.

App icons generated with ICONS REDMOND Z-Image Turbo

Trigger Words

Use both trigger words in your prompt:

ICONS ICREDM, a mobile app icon for a music player, flat design, purple gradient

The model responds to descriptions of app categories (fitness, weather, music, social, etc.) combined with color and style directives.

More app icon examples

Parameter | Value

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

Base model | Tongyi-MAI/Z-Image Turbo Trigger words | ICONS ICREDM Output size | 1024x1024 Inference steps | 20-30 Guidance scale | 7-9 Format | Safetensors

Setup

  1. Download the Safetensors file from HuggingFace or Civitai.
  2. Place it in your ComfyUI models/loras/ directory.
  3. Add a LoRA node and select the weights.
  4. Include ICONS ICREDM in your positive prompt.

Diffusers

Load the LoRA weights using load_lora_weights() from the HuggingFace diffusers library, then activate with set_adapters(). Specify Z-Image Turbo as your base pipeline.

Credits

GPU time sponsored by Redmond.AI. Licensed under Apache 2.0.

Common Issues

Problem | Cause | Fix

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

No icon style applied | Missing trigger words | Add ICONS ICREDM to prompt Low detail | Too few steps | Increase to 25-30 steps Wrong aspect ratio | Non-square resolution | Use 1024x1024 Poor colors | Low guidance scale | Try 7.5-9.0 CFG

FAQ

Q: Can I use this for commercial projects? A: Yes, Apache 2.0 allows unrestricted commercial use.

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

Q: What's the file size? A: 25.96 MB in Safetensors format.

Sources