🔥 HonōKitsune Ai — LoRA v1.0 🔥
(FR ⬇️ / EN ⬆️ — deux langues, une même flamme)
🇬🇧 EN — What does this model do?
HonōKitsune Ai is a lightweight LoRA (dim 128 / α ??) trained on 60 curated shots of my virtual VTuber fox-girl. Plug it into Stable Diffusion 1.5 (tested with A1111 / ComfyUI) and watch seven fluffy tails, vivid orange gradients and techno-miko fashion blaze onto the canvas.
Trigger word:
hono_kitsune_aiStyle: semi-realistic anime, punchy colours, warm glow
Strength: 0.65 – 0.9 (portrait) · 0.55 – 0.75 (full-body)
Works with: Any anime checkpoint (AbyssOrangeMix3, Anything-v5, Counterfeit-v3…)
Sample prompthono_kitsune_ai, close-up, gentle smile, long fiery orange hair, amber eyes, seven fluffy tails softly blurred, techno-miko outfit, warm mystical lighting, anime semi-realistic, high detail
Negative prompt: text, watermark, extra limb, worst quality, lowres
📝 License & credits
CC-BY-NC 4.0 — non-commercial use, credit “HonōKitsune Ai Project”.
Original character © 2025 InariVerse.
Chaque mot est une flamme, chaque image une étincelle — faites briller Honō ! 🔥🦊
Description
🔥 HonōKitsune Ai — LoRA SDXL 1.0 (v1.0) 🔥
🇬🇧 Version changes / notes (SDXL 1.0 port)
Re-trained from scratch on SDXL 1.0 base model (not merged / no DreamBooth conversion).
Image set expanded : 80 imgs @ 1024 × 1024
30 close-up · 25 half-body · 25 full-body.
Higher network capacity
network_dim 192 · network_alpha 96(≈ 65 MB).
Optimiser & schedule
Adafactor · LR 8e-4 (UNet) / 2e-5 (CLIP-G) · cosine with restarts × 4.
Training tricks
Min-SNR γ = 5, noise offset 0.05, flip aug., colour jitter light.
Bucketed aspect ratios up to 1152 px → handles vertical character shots better.
Recommended LoRA weight
0.55–0.7 portraits · 0.65–0.85 full-body (SDXL needs a bit more).
Trigger unchanged →
hono_kitsune_ai.Works best on Checkpoint “SDXL-TurboMix” / “Juggernaut XL” but any SDXL 1.0 compliant base is fine.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.