✨ A Pony-native LoRA for hyper-muscular men in glossy black latex — retrained from scratch for photorealistic Pony workflows.

💪 Trained on 49 curated high-resolution images of massive, bodybuilder-physique men in skintight black latex catsuits: mirror-shine latex, dramatic rim light, and sharp muscle definition. This V3 is a full retrain on CyberRealisticPony V18.0, not a port of the SDXL version — use it on realistic Pony XL checkpoints.

🎯 Trigger word: blacklatexmale

⚙️ Recommended settings

- Base: CyberRealisticPony V18.0 or any compatible realistic Pony XL checkpoint

- Strength: 0.65–0.85 (sweet spot 0.75 with the Epoch 10 file)

- CLIP Skip: -2

- CFG: 4.5–5.5

- Sampler: DPM++ 2M SDE + Karras (crisp photographic skin) or Euler a

- Prompt prefix: score_9, score_8_up, score_7_up, blacklatexmale

🎭 FaceDetailer tip

The training set leans hard on high-gloss latex and dramatic torso lighting. For close-up faces, run the FaceDetailer pass through the clean base model, or keep the LoRA at 0.0–0.25 inside the face mask only — otherwise the latex look can bleed into skin pores.

🧪 Training

- kohya-trainer (sdxl_train_network.py), FP16, 2× T4

- LoRA dim 64 / alpha 32, AdamW, lr 2e-4 cosine, 100 warmup steps

- 10 epochs, 4 repeats, 49 images (196 samples/epoch), max_token_length 225, keep_tokens 6

🖼️ Showcase images are AI-generated with this LoRA. All characters are fictional adults.

🆕 V3 — Pony Realistic Retrain

Full retrain from scratch on CyberRealisticPony V18.0 F16 — this is not the SDXL v1/v2 weights. Sharper photorealistic skin/latex contrast, cleaner specular highlights and rim light, tighter muscle definition, and less of the semi-2D / painterly smoothing from earlier Pony experiments. Pose control and tag adherence from Pony XL are preserved.

- 🎯 Trigger: blacklatexmale (only)

- 💪 Strength 0.65–0.85, sweet spot 0.75 (Epoch 10)

- 📁 Epoch 6 and Epoch 8 included for lighter styling / multi-LoRA stacking

- ⚙️ CLIP Skip -2, CFG 4.5–5.5, DPM++ 2M SDE + Karras recommended

- 🎭 FaceDetailer: keep LoRA at 0.0–0.25 on the face mask, or use the clean base for faces

Known limitations — faces & anatomy (and how to fix them)

Pony's base model leans painterly, so faces and genitalia can come out soft or smeary straight out of the base generation. That's the base model, not this LoRA — Pony was never built for photorealism. Please don't judge the LoRA on a single-pass render; run the second pass below and it cleans up:

- Faces: run a FaceDetailer / ADetailer pass after the base image, routed through the clean base model — or keep this LoRA at 0.0–0.25 inside the face mask only. Let the LoRA handle the body and the latex; let the base model handle the skin and pores.

- Genitalia: use masked inpainting with a local Detailer, and anchor the Detailer prompt to the body — one penis only, erect penis, emerging from the latex crotch opening, attached to the pelvis, pointing downward, testicles below the shaft base. (The old three-word version — penis, large penis, erection — gives the Detailer nothing to attach to.) Keep penis terms out of the global negative; they belong in the local box only.

- If the base image fuses anatomy into one blob, no Detailer will split it — regenerate the base until the forms are separated, then inpaint. For two figures, generate a good single-penis base first, then masked-inpaint the second one beside it, one penis per step; the mask decides the placement.

Rule of thumb: the LoRA does figure structure, latex texture and pose. Faces and explicit anatomy are separate local passes, done after.