just1n (Pony V3) is a character LoRA trained on Pony Diffusion XL (SDXL 1.0) for the trigger word just1n. It was built from 264 captioned reference images using Civitai’s Spine Controller and ai-toolkit, with a 32-rank LoRA at 1024px resolution, cosine learning-rate schedule, Adafactor optimizer, and joint UNet + text-encoder training. The run targeted a consistent male character read — mature, bara-leaning, realistic proportions, East Asian features, and recurring tags like short dark hair, glasses, facial hair, and an everyday “married father” look in casual settings.Fifteen epoch checkpoints are available from this training job (the full 20-epoch run did not complete). Sample generations skew toward upper-body and full-body portraits with natural lighting, outdoor or indoor photo backgrounds, and a range of body types from plump to muscular. Use just1n in your prompt at modest strength (roughly 0.7–1.0 to start) alongside quality tags; the bundled negative prompt from training (bad quality, low quality, jpeg artifacts, blurry, watermark, etc.) is a good baseline. Epoch 10–15 is usually a sensible starting range for Pony XL, but compare sample images across epochs to pick the best fit for your workflow.

redo-just1n

Model namejust1n (Ernie)Type / categoryCharacter · LoRATrained words (activation)just1n───Model description (main page — 2 paragraphs)just1n is a character LoRA trained on ERNIE-Image for the trigger word just1n. It targets a recurring male character with a realistic, bara-leaning look: mature build, East Asian features, short dark hair, glasses, facial hair, and an everyday “married father” vibe in casual indoor/outdoor settings. Training used 264 captioned images via Civitai Spine Controller and ai-toolkit, with a rank-32 LoRA at 512px, cosine LR schedule, AdamW 8-bit optimizer, and joint UNet + text-encoder training over a full 20-epoch run.Use just1n at roughly 0.7–1.0 strength on ERNIE-Image only (not Pony, Illustrious, or generic SDXL). Start around epoch 15–20 if you have multiple checkpoints; compare sample images across epochs before picking a favourite. Recommended negative: bad quality, low quality, worst quality, jpeg artifacts, blurry, pixelated, out of focus, watermark, text, signature.

Files with SHA256: 286cc3e8ba1c9493f4a0281c2e5b66a71b18149400f16899941125f575b98298