🌌 Overview
This is the V3 release (or Ernie V2) of my Loona LoRA from Helluva Boss. This version is the result of intensive refinement and a complete overhaul of my training process to provide the best possible experience for fans of the character.
🧩 Recommended Settings
* Trigger word: hbl00n4
* Strength: 0.8 - 1.2 (Experiment to find the "sweet spot" for your specific setup)
* Resolution: 1024x1024 / 1344x1344
* Samplers: LMS, DPM++ 2M, or Euler Ancestral.
🚀 Advanced Setup (Recommended)
For optimal results, I highly recommend using this LoRA in combination with the Turbo LoRA.
Suggested stack: Loona LoRA (1.0) + Turbo LoRA (0.8) + CFG (1.4)
💡 Pro Tip
The model is now highly responsive to prompts. Use:
hbl00n4, [your prompt here]
📝 Creator notes:
* Refined Training: I spent the majority of the last four days perfecting my training process and experimenting with different configurations to push this model to its limit.
* Qualified Dataset: I completely redid the dataset. It is now highly curated and qualified, resulting in a model that follows prompts accurately and performs effectively across various scenarios.
* Consistency: This version is much more reliable than previous iterations, offering cleaner lines and better character adherence.
⚠️ Known Limitations & Recommendations
* Anatomical Variations: You may occasionally encounter minor anatomical errors. These are usually tied to the specific seed or sampler being used rather than the model itself.
* Troubleshooting: If you encounter distortions, try changing the seed or switching between LMS, DPM++ 2M, or Euler Ancestral.
* Flexibility: While I have recommended settings, I encourage you to experiment with the strength (**0.8 to 1.2**) to see what works best with your preferred checkpoints and VAEs.
👤 Contact & Support
💬 Discord: hex_duck
💖 Brazilian Discord server: https://discord.gg/Q29RtePts5
If you enjoy this work and the effort put into this V3 update, consider supporting the creator:
☕ https://ko-fi.com/retrowavex32
Description
V1
* initial release
FAQ
Comments (3)
Is this for Qwen Image 2512 or a different version? It does work with QI 2512, but when generating I keep getting a human version rather than an anthropomorphic wolf unless I add "anthropomorphic" to the prompt and even then it doesn't look quite right. The images always have a realistic tone to them even if using keywords like "High contrast, stylized digital illustration, cool vibe, trending on pixiv, detailed shading"
This LoRa was created using standard QWEN image (not 2512). I even tried porting it to 2512, but the result was unsatisfactory.
@RetroWaveX It does work with QI 1.0, but I've stopped using that since 2512 has much better prompt adherence. For what it's worth there is another loona lora on this site for QI 2512 which works well. I just need to lower the strength a bit.