UI はローカライズ済みです。スキルの紹介と説明文書は選択した言語へ AI が自動翻訳し、コード・コマンド・パス・専門用語は原文のままです。原作者の文書が優先されます。
スキルを探す
生産性・共同作業 · SKILL.md

paired-inference

paired-inference

Route Pix2Pix-Turbo paired image-translation inference for edge-to-image, stochastic sketch-to-image, custom Pix2Pix checkpoints, and local paired Gradio demos.

Independent GitHub snapshot · Automated classification · Original documentation governs

スキルの説明

GitHub

name: paired-inference description: "Route Pix2Pix-Turbo paired image-translation inference for edge-to-image, stochastic sketch-to-image, custom Pix2Pix checkpoints, and local paired Gradio demos." disable-model-invocation: true metadata: disco-role: operating root-skill: img2img-turbo license: MIT

Paired Inference Router

Use this sub-skill when the task is Pix2Pix-Turbo paired image translation:

  • Canny edge-to-image from an RGB image plus a text prompt.
  • Stochastic sketch-to-image from a sketch plus a text prompt, guidance value, and seed.
  • Inference from a custom Pix2Pix-Turbo checkpoint produced by the paired training workflow.
  • Local Gradio demos for the paired Canny and sketch interfaces.

Do not use this sub-skill for unpaired CycleGAN-Turbo translation; route that to unpaired-inference. Do not use it for dataset layout, training, or checkpoint creation; route those to training.

Operating references

  1. Read CLI and API reference for exact selectors, flags, constructor and forward signatures, output files, and checkpoint schema expectations.
  2. Use workflows for ready-to-adapt edge, sketch, custom-checkpoint, and Gradio recipes.
  3. Use troubleshooting for CUDA/download, selector, prompt, resize, Canny/gamma, and Gradio failures.

Bundled safe helpers

These helpers do not run model inference or launch a server by default:

Safety and evidence limits

The source paired inference code is CUDA-oriented and constructs Stable Diffusion Turbo components plus Pix2Pix-Turbo LoRA checkpoints before inference. This generated sub-skill preserves the verified command/API behavior and safe parser/preprocessing helpers, but it does not claim that full model inference, checkpoint download, Gradio serving, or training was executed during skill drafting.

保存したスキル