ArtificialGuyBR

Home / Blog / Training LoRAs

IconsRedmond SDXL LoRA: Generate Flat App Icons with One Trigger Word

Generate consistent, Redmond-style app icons for iOS, macOS, and Windows using the IconsRedmond SDXL LoRA. Trigger word icredm, works at 1024x1024.

By ArtificialGuyBR Published Updated 2 sources cited Training LoRAs

TL;DR

  • Drop-in LoRA for SDXL that generates clean, flat app icons with the trigger word icredm
  • Best at 1024×1024 with weight 0.7–0.8; add minimalism or detailed to steer style
  • Trained on SDXL Base 1.0 with GPU support from Redmond.AI; licensed CreativeML OpenRAIL-M

What This LoRA Does

IconsRedmond is a specialist LoRA for Stable Diffusion XL that produces app-icon grids in a consistent "Redmond" flat style — the clean, rounded-corner aesthetic used by modern iOS, macOS, and Windows system icons. One trigger word (icredm) gives you coherent icon sets across weather, calculator, settings, camera, music, and dozens of other categories.

Grid of flat app icons in Redmond style showing weather, calculator, settings, and camera icons

The model was fine-tuned on stabilityai/stable-diffusion-xl-base-1.0 with compute sponsored by Redmond.AI. It's a focused tool: it does one thing (flat app icons) and does it consistently.

How to Use It

Trigger word: icredm (this was the instance prompt during training)

Recommended settings:

  • Resolution: 1024×1024 (SDXL native)
  • LoRA weight: 0.7–0.8 (author suggests testing 0.7 and 0.8)
  • Sampler: Euler a / DPM++ 2M Karras (standard SDXL samplers work)
  • Steps: 20–30
  • CFG: 5–7

Optional style tags:

  • ios icon app — pushes toward iOS/macOS system icon look
  • dribbble — slightly more polished, "designed" feel
  • minimalism — flatter, simpler shapes
  • detailed — more internal detail, less flat

Example prompt:

icredm, ios icon app, minimalism, weather app icon, blue gradient background, white glyph, rounded corners

Another icon grid showing music, photos, messages, and maps icons in consistent style

ComfyUI / A1111 / diffusers

ComfyUI: Load as a standard LoRA node, set strength_model to 0.7–0.8.

Automatic1111/WebUI: Place in models/Lora/, enable in the LoRA tab, set weight to 0.7–0.8.

diffusers (Python):

from diffusers import StableDiffusionXLPipeline
import torch

pipe = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16,
    variant="fp16"
).to("cuda")

pipe.load_lora_weights("artificialguybr/IconsRedmond-IconsLoraForSDXL")
pipe.set_adapters(["default"], adapter_weights=[0.8])

image = pipe(prompt="icredm, ios icon app, minimalism, calculator icon").images[0]

What the Images Actually Show

The sample images confirm the model delivers on its promise: grids of 12–20 icons per image, each icon internally consistent — same corner radius, same stroke weight, same visual language. Colors vary per icon (blue for weather, green for messages, orange for camera) but the palette feels curated. No photographic realism, no 3D render artifacts — just clean vector-style flats.

Close-up icon grid showing camera, photos, music, and settings icons with consistent styling

Known Limitations

  • Not a general-purpose LoRA — it only does flat icons
  • Sometimes needs 2–3 generations to get a perfect grid (author notes this)
  • Works best with simple prompts; complex scenes break the icon consistency
  • The Civitai page hosts a different model (SD 1.5 version for LiberteRedmond), not this SDXL variant

License

CreativeML OpenRAIL-M — same as the SDXL base model. Commercial use permitted under the license terms.

Common Errors and Fixes

Error | Cause | Fix | Source

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

Icons look photographic or 3D | Weight too high or wrong base model | Use weight 0.7–0.8 on SDXL Base 1.0 only | HF model card Inconsistent corner radius / style | Prompt too complex | Simplify prompt; use only icredm + 1–2 style tags | Author notes Blurry / low detail | Resolution too low | Generate at 1024×1024 native SDXL resolution | HF model card

FAQ

Q: Is this the same as the IconsRedmond on Civitai? A: No. The Civitai model (ID 206191) is Icons.Redmond 1.5V for SD 1.5 (LiberteRedmond). This HF model is for SDXL.

Q: What's the difference between V1 and V2? A: V2 (separate HF repo) has further training improvements. V1 is the original release documented here.

Q: Can I use this commercially? A: Yes, under CreativeML OpenRAIL-M — same license as SDXL Base.

Q: Does it work with SD 1.5 or Pony Diffusion? A: No. It's trained specifically for SDXL Base 1.0.

Sources

  • Hugging Face model card: https://huggingface.co/artificialguybr/IconsRedmond-IconsLoraForSDXL
  • Civitai page (SD 1.5 variant): https://civitai.com/models/206191
  • Author's demo space: https://huggingface.co/spaces/artificialguybr/artificialguybr-demo-lora
  • Redmond.AI (GPU sponsor): https://redmond.ai/