Skip to content
Back to search

PocketMen-with-you

Create high-fidelity Codex companions from 2+ photos with local open-weight neural editing and no OpenAI API key.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/six-nut-pocketmen-with-you/badge.svg)](/listings/six-nut-pocketmen-with-you)

HTML

<a href="/listings/six-nut-pocketmen-with-you"><img src="/api/listings/six-nut-pocketmen-with-you/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Upload at least two images and ask:

Use $pocketmen-with-you to create and install a companion from these references.
Use Neural Local Studio when my hardware supports it.
Style: hero-chibi.
Subject type: person.
Quality: max.
Preserve identity strictly and do not use hatch-pet or any OpenAI API key.

The skill should visually summarize stable identity traits from the references and pass them through `--identity-notes`.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 177 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <div align="center">
  2. <img src="assets/logo-512.png" width="132" alt="PocketMen with You companion capsule logo" />
  3. <h1>PocketMen with You</h1>
  4. <p><strong>Turn 2+ photos into a high-fidelity Codex companion — locally.</strong></p>
  5. <p><strong>v0.3: Neural Local Studio. No hatch-pet. No OpenAI API key. Open-weight multi-reference editing when your GPU can run it.</strong></p>
  6. <p>
  7. <img src="https://img.shields.io/github/stars/six-nut/PocketMen-with-you?style=for-the-badge&color=yellow" alt="GitHub stars" />
  8. <img src="https://img.shields.io/github/issues/six-nut/PocketMen-with-you?style=for-the-badge" alt="GitHub issues" />
  9. <img src="https://img.shields.io/github/v/release/six-nut/PocketMen-with-you?style=for-the-badge" alt="Release" />
  10. </p>
  11. <p><a href="https://github.com/six-nut/PocketMen-with-you/releases/tag/v0.3.0"><strong>Try now</strong></a> · <a href="https://github.com/six-nut/PocketMen-with-you"><strong>⭐ Star</strong></a> · <a href="https://github.com/six-nut/PocketMen-with-you/discussions/11"><strong>Share example</strong></a></p>
  12. </div>
  13. [简体中文](README.zh-CN.md)
  14. ## Why v0.3 exists
  15. The deterministic v0.2 engine was reliable and private, but it could only warp supplied pixels. It could not genuinely invent a new waving hand, a natural running stride, a faithful kitten pose, or a polished 3D chibi redesign.
  16. v0.3 adds a **real local generative engine**. When supported hardware is present, PocketMen uses an open-weight multi-reference image editor to create a canonical master and state-specific animation key poses, then applies deterministic micro-motion, atlas assembly and QA. The result is far closer to a modern hosted image editor while remaining independent from OpenAI ImageGen.
  17. ## Engine stack
  18. ```text
  19. 2+ reference images
  20. ↓
  21. Codex visual identity lock
  22. ↓
  23. PocketMen hardware doctor
  24. ↓
  25. ┌─────────────────────────────────────┐
  26. │ Neural Local Studio │
  27. │ default: FLUX.2 [klein] 4B │
  28. │ optional: Qwen-Image-Edit-2511 │
  29. │ multi-reference identity editing │
  30. └─────────────────────────────────────┘
  31. ↓
  32. canonical master + semantic key pose per state
  33. ↓
  34. flat chroma extraction / alpha cleanup
  35. ↓
  36. deterministic micro-animation
  37. ↓
  38. 8 × 9 transparent Codex atlas
  39. ↓
  40. validation + contact sheet + GIF previews
  41. ↓
  42. pet.json + spritesheet.webp
  43. ↓
  44. Codex local install
  45. ```
  46. If the neural runtime or compatible GPU is unavailable, PocketMen falls back to the deterministic v0.2 engine instead of asking for `OPENAI_API_KEY`.
  47. ## Local neural backends
  48. ### FLUX.2 [klein] 4B — default
  49. - local text-to-image and image editing;
  50. - multi-reference input;
  51. - Apache-2.0 weights;
  52. - suitable for consumer NVIDIA GPUs;
  53. - selected automatically on compatible hardware.
  54. ### Qwen-Image-Edit-2511 — Identity-Max
  55. - multi-image editing;
  56. - strong character consistency and viewpoint editing;
  57. - Apache-2.0;
  58. - heavier than the default backend;
  59. - install explicitly when identity fidelity matters more than setup size or latency.
  60. PocketMen intentionally does **not** make FLUX.2 [dev] the default because its model license is non-commercial.
  61. ## Quality profiles
  62. - `draft`: one neural canonical master, deterministic nine-state animation.
  63. - `balanced`: canonical master + state-specific neural key poses; left-running may derive from right-running to reduce drift.
  64. - `max`: canonical master + independently generated key pose for every state.
  65. ## Quick start in Codex
  66. Upload at least two images and ask:
  67. ```text
  68. Use $pocketmen-with-you to create and install a companion from these references.
  69. Use Neural Local Studio when my hardware supports it.
  70. Style: hero-chibi.
  71. Subject type: person.
  72. Quality: max.
  73. Preserve identity strictly and do not use hatch-pet or any OpenAI API key.
  74. ```
  75. The skill should visually summarize stable identity traits from the references and pass them through `--identity-notes`.
  76. ## CLI
  77. Hardware check:
  78. ```bash
  79. pocketmen doctor
  80. ```
  81. Create with automatic engine selection:
  82. ```bash
  83. pocketmen create \
  84. --reference photo-1.jpg \
  85. --reference photo-2.jpg \
  86. --name "Mochi" \
  87. --subject-type animal \
  88. --style soft-real \
  89. --engine auto \
  90. --quality balanced \
  91. --output ./pocketmen-output \
  92. --install
  93. ```
  94. Force the default neural model:
  95. ```bash
  96. pocketmen create ... --engine neural --backend flux2-klein-4b --quality max
  97. ```
  98. Use the heavier identity backend:
  99. ```bash
  100. pocketmen create ... --engine neural --backend qwen-image-edit-2511 --quality max
  101. ```
  102. ## One-click setup
  103. `INSTALL_SKILL.bat` installs the skill and creates an isolated runtime. On Windows, the installer checks NVIDIA VRAM; a compatible machine automatically receives the neural dependency profile.
  104. To explicitly prepare and cache the default local neural model:
  105. ```text
  106. PREPARE_NEURAL_ENGINE.bat
  107. ```
  108. For Qwen Identity-Max:
  109. ```text
  110. PREPARE_IDENTITY_MAX.bat
  111. ```
  112. Model weights are downloaded from their original model hosts into the normal Hugging Face cache; they are not committed to this repository.
  113. ## Codex pet contract
  114. PocketMen emits an 8-column × 9-row atlas, 192×208 per cell, 1536×1872 overall. States are:
  115. `idle`, `running-right`, `running-left`, `waving`, `jumping`, `failed`, `waiting`, `running` (Codex working, not physical running), and `review`.
  116. ## What “ImageGen-like” means here
  117. PocketMen narrows the quality gap specifically for **reference-driven Codex companion creation** by combining strong open-weight image editing, multi-reference identity anchors, structured prompts, dedicated state generation, deterministic alpha extraction and strict atlas QA.
  118. It does **not** claim bit-for-bit or general-purpose parity with a proprietary frontier model. Broad world knowledge, typography, complex scene reasoning and difficult multi-object composition can still differ. For the narrow “2+ photos → consistent animated companion” workflow, however, v0.3 is designed to use the strongest local path available before falling back to simple warps.
  119. ## Privacy
  120. Raw personal references remain local by default and are never copied into public Git commits/releases. The final install package contains only `pet.json` and `spritesheet.webp`.
  121. ## Community & maintenance playbook
  122. For operation cadence (24h follow-ups, community sync templates, and weekly reports), use:
  123. - `docs/maintenance/promo-cycle-playbook.md`
  124. ## License
  125. PocketMen code: MIT.
  126. Optional model weights retain their own licenses. The default FLUX.2 [klein] 4B and Qwen-Image-Edit-2511 backends are referenced, not redistributed.
  127. PocketMen with You is not affiliated with Nintendo, The Pokémon Company, Game Freak, OpenAI, Black Forest Labs, Alibaba/Qwen, Hugging Face, or GitHub.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.