ArtificialGuyBR

Home / Blog / Training LoRAs

PixelArtRedmond LoRA for SD 1.5: Trigger Words

PixelArtRedmond is an SD 1.5 LoRA for pixel art: the trigger words, recommended weights, why Liberte is the base, and ComfyUI settings.

By ArtificialGuyBR Published Updated 2 sources cited Training LoRAs

TL;DR

  • LoRA for SD 1.5 that generates consistent pixel art using triggers "Pixel Art, PixArFK"
  • Trained on Liberte model with Redmond.AI GPU sponsorship
  • Works best with Pixel Art Extension + 4x/8x downscaling

What This Model Does

PixelArtRedmond 1.5V is a Low-Rank Adaptation (LoRA) for Stable Diffusion 1.5 that specializes in generating pixel art style images. It was trained on the Liberte model (a SD 1.5 fine-tune) using a large dataset of pixel art images. The training compute was sponsored by Redmond.AI.

The model excels at producing clean, consistent pixel art across various subjects — characters, scenes, game assets, and even coloring book style line art. Sample outputs show astronauts, fantasy characters, animals, and everyday scenes all rendered in a cohesive pixel art aesthetic.

Astronaut in space pixel art Astronaut in space — demonstrates the model's ability to handle sci-fi subjects in pixel style

Trigger Words and Usage

Two trigger words are required to activate the pixel art style:

  • Pixel Art — primary style trigger
  • PixArFK — secondary trigger (instance prompt used during training)

Include both in your prompt:

Beautiful blonde girl, stunning, detailed, portrait, pixel art, PixArFK,
  • Base model: Liberte (SD 1.5) or any SD 1.5 checkpoint
  • LoRA weight: Start at 0.8–1.0, adjust to taste
  • Extension: Use Pixel Art Extension for ComfyUI/Automatic1111 for cleaner pixel-perfect output
  • Downscale: Generate at higher resolution, then downscale by 4x or 8x for authentic pixel density
  • Sampler: Euler a or DPM++ 2M Karras work well
  • Steps: 20–30
  • CFG: 7–9

Super cat hero pixel art Super cat hero — shows character consistency and style coherence

Training Details

  • Base model: Liberte (SD 1.5 derivative)
  • Training data: Large pixel art dataset
  • Instance prompt: "Pixel Art"
  • Compute sponsor: Redmond.AI
  • License: Bespoke LoRA Trained License — commercial use allowed (rental), derivatives allowed, no credit required

Example Prompts

Subject | Prompt

|---|---

Portrait | Beautiful blonde girl, stunning, detailed, portrait, pixel art, PixArFK Game asset | A warrior, game asset, detailed, pixel art, PixArFK Scene | A highly detailed mother in her kitchen preparing meal, window sunny day, happy, smiling, pixel art, PixArFK Creature | A cute dinosaur, pixel art, PixArFK Sci-fi | An astronaut on space, pixel art, PixArFK Cute dinosaur pixel art Cute dinosaur — demonstrates the model's range on creature designs

Common Errors and Fixes

Error | Cause | Fix | Source

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

Blurry/soft pixels | No downscaling applied | Downscale output 4x or 8x after generation | Model card Style not activating | Missing trigger words | Include both "Pixel Art" and "PixArFK" | Model card Inconsistent pixel density | Base model mismatch | Use Liberte or standard SD 1.5 checkpoint | Model card

FAQ

Q: Does this work with SDXL? A: No, this LoRA is trained specifically for Stable Diffusion 1.5 (via Liberte model).

Q: Can I use this commercially? A: Yes, the Bespoke LoRA Trained License permits commercial use (rental), derivatives, and requires no credit.

Q: What's the Pixel Art Extension? A: A ComfyUI/Automatic1111 extension that post-processes generations for cleaner pixel-perfect output. The author recommends it for best results.

Q: Why downscale by 4x or 8x? A: SD generates at latent resolution; downscaling creates authentic pixel density and removes soft edges.

Sources


Model by ArtificialGuyBR. Explore more at artificialguy.com.