Updated for Z-Image

The one bar prison updated for Z-Image Turbo. This is an initial re-train under Zimage using the same data-set as Pony but re-captioned to mesh better with Z-Image. resolutions up to 1536px in the AI Toolkit training with 4000 steps.

Use strengths between 0.7 and 1 depending. If you have other LoRAs you almost certainly need to tone it down.

I can't wait to see what you make with this.

Main trigger: 0n3barpr1s0n

Additional triggers:

metal collar

wearing heels

barefoot

tiptoe

ankle cuffs

I think this can be more flexible and I'll try to do more. But enjoy what it is for now.

Update Pony version 2.0

Finally got this working. Pony was assuming anything with "pole" referred to a stripper pole so I have changed the trigger to 0n3barpr1s0n which has completely solved the issues I had.

The new PonyXL trained version is tested working across photo realistic and anime models.

Optional tags: cuffs, tiptoe, locked, metal-collar

I typically use the following rough txt2img settings:

Steps: 40,

Sampler: DPM++ 3M SDE,

Schedule type: Exponential,

CFG scale: 4,

Size: 1200x1200,

Model: realDream_sdxlPony12

Sd1.5 version 1.0:

Primary: standing on a pole stand

Also: With a collar, heels, and handcuffs

Trained on a combination of my own 3D renders and actual photographs ~ 650 images total and against the base SD 1.5 model to maximize compatibility. Sample images have prompt info - You may want to start there but the primary trigger is "on a pole stand" (using one bar prison in the prompt will confuse SD into thinking there is some kind of cage)

Weight between 0.6 and 1 depends a lot on the base checkpoint. (0.6 - 0.8 for photorealistic or 0.8-1.0 for anime seems to work well)

Please let me know what you think, suggestions and otherwise. I hope to do more!