Trained on official in-game model + high-quality shots from the "Forged in Fury" cinematic for accurate colors, lighting, and details.
How to Use It Effectively
Trigger Word: driftpixel_shyvana
⚠️ Important: Do not use the trigger word by itself — results will be half-baked. Always combine it with detailed descriptions of pose, expression, outfit, lighting, and mood for the best quality.PICK THE RIGHT VERSION!
v1.0+ – Illustriousv2.0 – PonyRecommended Settings
LoRA Weight: 0.9 – 1.2
Sampler: Euler a (recommended) Alternative: DPM++ 2M Karras
Scheduler: Normal or Beta 1.1
Steps: 20 – 40
CFG Scale: 5 – 7
Resolution: 768×1024 (great starting point) I often use 832×1216
Recommended Universal Negative Prompt:
low quality, worst quality, blurry, jpeg artifacts, bad anatomy, bad hands, deformed, distorted, disfigured, extra limbs, missing limbs, malformed limbs, fused fingers, too many fingers, mutated hands, poorly drawn hands, poorly drawn face, long neck, gross proportions, cropped, out of frame, watermark, text, signature, logo, usernameBonus for ComfyUI Users
A full ComfyUI workflow is embedded in the metadata of every sample image I uploaded. Just download any showcase image and drag & drop it into ComfyUI — it will load the complete workflow automatically.This LoRA represents many hours of careful training and testing. If you find it useful, please 👍 like, 💬 leave a comment with your results, and feel free to ⚡ tip. it really helps support future projects!
Description
Base model: Illustrious Strange Seduction
Status: Bad version – do not use
This was my first public LoRA release. Unfortunately it ended up significantly overcooked during training.
As a result, it suffers from:
inconsistent poses and anatomy
very limited flexibility in style changes
poor response to dynamic / complex poses
For these reasons this version is not recommended and should be skipped.
FAQ
Comments (2)
Was very fast, nice. It's a bit iffy forcing the style but that's understandable given the lack of source material which forces use of in-game models.
For people trying to use I found using it with a different style lora and scheduling it to be weaker in the second half of generation like so: "[<lora:Shyvana_DriftPixel_v1.0:1>:<lora:Shyvana_DriftPixel_v1.0:0.2>:0.5]"
helps immensely.
Would be cool to see an update when there's more material available for training.
Hey, i updated my LoRA and i think i manage to make it slightly more manoeuvrable. Do you mind giving it another shot ?