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.
Description
For your Pony V3 upload (redo-just1n, checkpoints in downloads/pony/redo-just1n/V3_3078747/), CivitAI splits this into two fields with different jobs.
Version name (short label)
Keep it scannable — base model + which checkpoint you're publishing:
• Pony V3 — epoch 15 (if uploading the final available checkpoint)
• Or Pony V3 — epochs 10–15 available if you're publishing one file but noting alternatives exist
Avoid putting hyperparameters or trigger words here; those belong in notes.
Version notes (what users need to run it)
This is where the operational detail goes. For your Pony run, include:
Must-have
• Trigger word: just1n
• Base model: Pony Diffusion V6 XL (SDXL 1.0 family) — trained against Civitai checkpoint 257749@290640
• Recommended strength: start ~0.7–1.0
• Clip skip: 1
• Recommended epoch: you only have epochs 1–15 (job failed before 20); say which you uploaded and that 10–15 is a reasonable starting range
• Resolution: trained at 1024px
Training summary (builds trust)
• Character LoRA, 264 images, rank 32 / alpha 32
• Engine: Civitai Spine Controller + ai-toolkit
• Optimizer: Adafactor, LR 0.0005, cosine schedule, text encoder trained
Honest caveats (important for this version)
• Training job failed after 15 epochs (planned 20) — mention possible under/overfitting vs a full run
• Pony-only — won't behave correctly on SDXL generic, Illustrious, Ernie, etc.
• Suggest Pony quality tags (score_9, score_8_up, etc.) and your negative prompt baseline
Optional but useful
• Link to other versions if you publish them (V2 Ernie, V4 Illustrious) so people don't mix bases
• Sample prompt that worked in training (one clean example is enough)
• "Not included: full 20-epoch set; epochs 16–20 unavailable"