| --- |
| license: other |
| license_name: creativeml-open-rail-mop-attachment-b |
| license_link: https://civitai.com/models/license/2811401 |
| pipeline_tag: text-to-image |
| tags: |
| - coreml |
| - sdxl |
| - lcm |
| - dmd |
| - localmuse |
| --- |
| |
| # MoP Insta v1.0 DMD for LocalMuse |
|
|
| Core ML conversion of **MoP (Mixture of Perverts) DMD — Insta v1.0 DMD** by GBRX, from [Civitai model version 2811401](https://civitai.com/models/1854124?modelVersionId=2811401). This repository contains the compiled on-device inference files used by LocalMuse, not the original checkpoint. |
|
|
| ## Generation profile |
|
|
| - SDXL text-to-image at **1024 × 1024** (fixed Core ML input shape) |
| - **LCM** sampler/scheduler, **10 steps** by default; author recommends 6–14 steps, with 8–12 suggested for this version |
| - **CFG 1.0** by default; author recommends 1.0–1.3 |
| - Optional LocalMuse Face Detail can refine a detected face using the included VAE Encoder |
| - Minimum supported device class: **8 GiB RAM** |
|
|
| The author's example PNG uses an LCM sampler, 10 steps, CFG 1.0, 1152 × 1536, and ComfyUI `ddim_uniform` schedule. Its workflow also applies a UNet downscale patch, two FaceDetailer passes, and film grain. This Core ML artifact targets 1024 × 1024 and uses LocalMuse's LCM implementation and Face Detail; outputs will differ from that ComfyUI workflow. |
|
|
| ## Components |
|
|
| - UNet: exact MoP weights, 6-bit k-means palettization, split into two Core ML chunks |
| - TextEncoder (CLIP-L): exact MoP conditioning layers, FP16. Six non-finite tensors in its unused final transformer layer were repaired from a pinned SDXL Base checkpoint; the used conditioning layers remain from MoP. |
| - TextEncoder2 (OpenCLIP-bigG): exact MoP weights, 8-bit palettization |
| - VAE Encoder and Decoder: exact MoP weights, Core ML conversion |
| - Tokenizer: pinned SDXL Base configuration accompanying the single-file checkpoint |
|
|
| See `PROVENANCE.json` for source hashes, converter versions, and validation details. Inference runs locally on the device. |
|
|
| ## License and attribution |
|
|
| The source model is licensed under [CreativeML Open RAIL++-M with the MoP Attachment B](https://civitai.com/models/license/2811401). The complete terms are in `LICENSE` and apply to this derivative. Attachment B restricts monetized image-generation services and sales of model weights or merges. Credit: **GBRX, MoP (Mixture of Perverts) DMD**. This conversion is independently distributed for free and is not endorsed by the original creator or Stability AI. |
|
|
