LP++ | PSX/PS1 Inspired

Hex_DuckHex_Duck
26
249
ps1/psxlow polystyleaesthetic
V3 - Ideogramv2ใƒปQWENv2v1
model-image-0
model-image-1
model-image-2
model-image-3
model-image-4
model-image-5
model-image-6

์„ธ๋ถ€ ์ •๋ณด

ํŒŒ์ผ ๋‹ค์šด๋กœ๋“œ (1)

๋ชจ๋ธ ์„ค๋ช…

๐ŸŒŒ Overview

LP++ (Low Poly++) is a LoRA inspired by the classic PS1 / PSX low poly aesthetic, but pushed further into a more refined and high-quality direction.

It is designed to generate artistically rich images with enhanced fidelity, blending the nostalgic charm of early 3D graphics with modern rendering consistency.


๐Ÿงฉ Recommended Settings (ERNIE)

  • Trigger word: ERPSX:

  • Strength: 0.8 - 1.2 (Experiment to find the "sweet spot" for your specific setup)

  • Resolution: 1024x102

  • Samplers: LCM, DPM++ 2M, or Euler Ancestral

๐Ÿงฉ Recommended Settings (QWEN-IMAGE)

  • Trigger word: ERPSX:

  • Strength: 0.8 - 1.3 (Experiment to find the "sweet spot" for your specific setup)

  • Resolution: 1344x1344

  • Samplers: DDIM, DPMPP_SDE, er_sde, UNI_PC, EULER

  • Scheduler: BETA, BETA57

๐Ÿงฉ Recommended Settings (IDEOGRAM)

  • Trigger word: ERPSX:

  • Strength: 0.8 - 1.3 (Experiment to find the "sweet spot" for your specific setup)

  • Resolution: 1344x1344

  • Samplers: Euler_ancestral, res_2m, euler, ddim

  • Scheduler: BETA, BETA57 OR Ideogram Scheduler

  • STEPS: Turbo (12), Balanced (28) or quality (48)


๐Ÿš€ Ernie recommended Setup
For optimal results with ernie, I highly recommend using this LoRA in combination with the Turbo LoRA.

Suggested stack: LP++ LoRA (1.0) + Turbo LoRA (0.8) + Cfg (1.4)

๐Ÿš€ QWEN recommended Setup
For optimal results with Qwen, I highly recommend using this LoRA in combination with the 8 Step Lightning LoRA.

Suggested stack: LP++ LoRA (1.0) + 8 Step Lightning LoRA (1.0) + Cfg (1.0)


๐Ÿ’ก Pro Tip
For best results Use:

ERPSX: PS1/PSX graphics, low resolution, low poly, pure 32-bit

[your prompt here]


๐Ÿš€ V2 Notes -

  • Simplified Prompting:
    This version introduces an improved captioning approach, allowing the LoRA to achieve the desired PSX-style output with significantly fewer tags. The goal was to reduce prompt complexity while maintaining visual quality.

  • Extended Training:
    V2 was trained for a longer duration, resulting in more stable outputs, better generalization, and improved consistency across different scenarios.

  • Refined Behavior:
    The LoRA is more predictable and responsive, making it easier to control composition and style even with simpler inputs.


โš ๏ธ Known Limitations & Recommendations

  • QWEN Polygonization Resistance: The Qwen version does an excellent job and captures the LP++ visual beautifully. However, the base model is somewhat resistant to โ€œpolygonizationโ€ (making things look properly blocky and low-poly). You must force it by following the prompt structure mentioned in the Pro Tips.

  • Using Character LoRAs with QWEN: When using character LoRAs alongside this one, the character LoRA will likely overpower the low-poly effect. It is highly recommended to lower the strength/weight of your character LoRA so the LP++ style can shine through.

Aside from these specific Qwen behaviors, the model is currently stable and performs as expected across tested scenarios.

๐Ÿ‘ค Contact & Support

๐Ÿ’ฌ Discord: hex_duck
๐Ÿ’– Brazilian Discord server: https://discord.gg/Q29RtePts5

If you enjoy this work and the effort put into this release, consider supporting the creator:

โ˜• https://ko-fi.com/retrowavex32

์ด ๋ชจ๋ธ๋กœ ๋งŒ๋“  ์ด๋ฏธ์ง€