DORO EPIC SKIES - Lightning
Versions
v1 |
DES_LIGHTNING_1- Initial release
Compatibility
Illustrious XL 🟢 full
Pony XL 🟡 partial
Quick Start
🏷️ Trigger: DES_LIGHTNING_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 = 0.5 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:
1.0–2.0- subtle atmosphere, soft glow, gentle storm hints2.5–3.5- full effect active: branching lightning, dramatic clouds, epic palette ⭐4.0+- overpowered, style dominates composition
Description
📸 Dataset: 15 hand-picked images - analog airbrush and acrylic illustrations from the 70s–90s depicting storms, lightning strikes, and dramatic skies. No photographs. Each image was selected for strong compositional contrast, authentic analog texture, and stylistic consistency.
✨ Emergent effects:
Branching lightning - detailed, glowing bolts with halos and natural spread patterns
Epic storm palette - deep blues, electric greens, vivid purples; saturated and cinematic
Volumetric clouds - massive, backlit cumulonimbus with soft analog gradients
Dramatic contrast - deep shadow regions opposing bright discharge zones, vintage illustration feel
⚠️ Side effect: At high weights (>3.5), lightning and storm elements may bleed into foreground subjects. Workaround: lower LoRA weight or use prompt weight to anchor the subject (character:1.3).
💡 Bonus use: Works as a visual FX layer - stack with character or object LoRAs to add atmosphere without overriding the primary subject. Great for title cards, posters, and fantasy scenes.
What happened under the hood
The dataset consisted entirely of non-photographic analog illustrations - airbrush and acrylic works created in the 70s–90s. This pushed the model to learn color relationships and light behavior from painted sources rather than photographic references, producing gradients and glow effects that feel hand-crafted rather than rendered.
Because the training set was compact (15 images, 300 steps, Dim 32), the LoRA operates as a weight bias rather than a style override. It nudges generation toward storm aesthetics - especially background and mid-ground elements - without displacing characters or primary subjects. The effect scales predictably with weight, making it easy to dial in.
Compatibility with Pony XL is a direct result of the training setup: the text encoder was disabled during training to conserve VRAM. This means the entire LoRA lives exclusively in UNet weights - there are no text encoder deltas to conflict with Pony's different tokenizer or embedding space. The visual concepts transfer cleanly across architectures. Expect slightly softer results on Pony; adjust weight upward by ~0.3–0.5 to compensate.
Analog Airbrush reference:
https://en.wikipedia.org/wiki/Airbrush
❤️ Artificial Inspiration by DORO