DORO EPIC COVERS - Iron Maiden 98 - Virtual XI
Versions
v1 |
DEC_VIRTUAL_1- Initial release.
Compatibility
Illustrious XL - 🟢 full
Pony XL - ❓ not tested
Quick Start
️ 🏷️ Trigger: DEC_VIRTUAL_1
⚠️ High-offset LoRA - effective range starts at 2.0+, not the usual 0.5-1.0. This is intentional: UNet-only training + Alpha/Dim = 1.0 scaling keep the signal quiet by design - preserving composition integrity, preventing anatomy bleed, and maximizing cross-model compatibility. Not broken. Just needs more pressure.
🏆 Sweet spot:
1.5 - 2.0- style bleeds in: oil painting texture, warm-cool contrast, painterly gradients2.0 - 2.5- full activation: Grant's palette, organic skin texture, dramatic lighting ⭐2.5+- overpowered, Eddie anatomy bleeds into any subject
Booster tags (if weight < 1.5):
Organic texture: exposed brain, veins, bulging muscles, textured skin
Face: skeletal grin, sinister smile, glowing red eyes, sharp teeth
Environment: burning cityscape, hellish atmosphere, orange flames, cyan rim lighting
Description
Style LoRA distilled from Melvyn Grant's 1998 oil painting - the cover of Iron Maiden's Virtual XI - capturing its painterly depth, spectral palette, and organic horror texture.
Dataset: 17 hand-curated and retouched fragments - sourced from a high-resolution production scan of the original artwork, oil and airbrush texture focus. The VR-boy fragment was excluded to keep the signal clean. Sub-scenes: Identity (Eddie's face, eyes, grin), Material (veined skin, exposed brain, blood beads), Motifs (clawed hand, anatomy), Environment (burning ruins, souls in flame).
✨ Emergent effects:
Organic surface shader - replaces standard textures with wet-flesh, open-brain and bulging-vein detail without explicit prompting
Spectral contrast - deep orange fire vs cyan-teal rim lighting locks in automatically, even on neutral subjects
Painterly depth - soft gradients and visible brushstroke density mimic Grant's oil-on-canvas technique
Eddie mode - semantic triggers
monster,demon,undead) activate the full mascot anatomy: rictus grin, distinctive skull, red eyes
⚠️ Side effect: Eddie anatomy can bleed into non-monster subjects at weights above 2.5 - skull structure, grin, vein texture may appear on human characters. Workaround: keep weight at 2.0-2.5 and avoid mascot-domain tags for clean portrait work.
💡 Bonus use: Apply as a color-grading and texture layer on top of any illustration style - the warm-cool spectral palette transfers independently of the horror content at lower weights.
What happened under the hood
The LoRA was trained UNet-only (TE LR = 0) with dim 16 in Juicer architecture - a narrow bottleneck that acts as a high-frequency filter, pushing the network to encode painterly density rather than raw pixel patterns. Triple suppression of the text encoder prevents the "sticker effect" and preserves Illustrious XL's compositional flexibility.
Training ran for 5,700 steps at a conservative LR of 1.5e-5 (energy 0.085) with cosine restarts over 3 cycles. Three spectral checkpoints were saved at ~1,900-step intervals, covering the full exposure range from soft style influence to hard mascot lock-in.
The high working range (2.0-2.5) is a direct consequence of the UNet-only architecture: without text encoder correction the signal stays quiet by design, requiring more pressure to activate - but that same quietness is what makes it compositable across models without anatomy bleed.
Virtual XI - Iron Maiden (Wikipedia):
https://en.wikipedia.org/wiki/Virtual_XI
Melvyn Grant (official site):
❤️ Artificial Inspiration by DORO