Skip to content
Back to search

tools

Tools, ComfyUI workflows and benchmark configs from a 4x RTX 3090 local-inference rig

Safety scan

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

Checked today

Check it deeper on your machine

Its scripts were only partly scanned.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/alesha-pro/tools with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/alesha-pro/tools'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/alesha-pro--tools/badge.svg)](/listings/alesha-pro--tools)

HTML

<a href="/listings/alesha-pro--tools"><img src="/api/listings/alesha-pro--tools/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

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 · 112 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. # tools
  3. Things I build while running models on my own hardware,
  4. cleaned up enough for other people to use.
  5. [![License](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
  6. [![Rig](https://img.shields.io/badge/measured%20on-4x%20RTX%203090-76b900.svg)](https://x.com/superalesha)
  7. [![X](https://img.shields.io/badge/@superalesha-follow-black.svg)](https://x.com/superalesha)
  8. </div>
  9. ---
  10. ## Sections
  11. Every section stands on its own, with its own installer and its own docs. There
  12. is no shared setup to get through first.
  13. ### [minimax-h3](minimax-h3)
  14. Video with sound, generated at home.
  15. [MiniMax H3](https://huggingface.co/Comfy-Org/MiniMax-H3) writes the picture and
  16. the synchronized audio in one pass, so voice, effects and music come out of the
  17. same generation as the frames. Open weights.
  18. Thirteen ComfyUI workflows in two sets. One set runs on any single 24 GB card. The
  19. other puts every card in a multi-GPU box to work and hits **3.12x on four
  20. cards**, taking a 15-second shot from 19 minutes of sampling down to 6:41.
  21. Ships with the custom nodes, the patch that makes the turbo LoRA actually load,
  22. and a three-command installer.
  23. ```bash
  24. cd minimax-h3 && ./install/setup.sh check
  25. ```
  26. ### [qwen-image-2.1](qwen-image-2.1)
  27. Image generation, editing and transparent PNGs on **one GPU**. The baseline is
  28. an RTX 3090 with 24 GB VRAM, using official INT8 ConvRot weights.
  29. Five native ComfyUI workflows, an isolated installer, SHA-256 model verification,
  30. and a script that submits a generation and checks the resulting PNG. Includes
  31. an [English guide](qwen-image-2.1/README.md), a
  32. [Russian guide](qwen-image-2.1/README.ru.md), and an
  33. [installation brief for an agent](qwen-image-2.1/INSTALL-WITH-AGENT.md).
  34. For smaller GPUs, use the **[low-VRAM guide](qwen-image-2.1/LOW-VRAM.md)**:
  35. GGUF Q4_K_M, CPU offloading, 7 workflows, and measured memory requirements.
  36. ```bash
  37. cd qwen-image-2.1
  38. python3 install/setup.py check
  39. ```
  40. ### [skills](skills)
  41. Agent skills that stand on their own.
  42. **[hand-drawn-canvas-animation](skills/hand-drawn-canvas-animation)** makes short
  43. films with authored poses and expressive strokes in JavaScript and Canvas 2D.
  44. It covers pencil, ink, risograph, screen print and drawings interacting with
  45. real photos, plus sand animation and projected paper. The
  46. [60-second phoenix film](skills/hand-drawn-canvas-animation/examples/becoming-phoenix/)
  47. combines the materials through one story, including a moving storm and wings
  48. that unfold from a book. The skill includes editable examples, original sound,
  49. an MP4 renderer and browser regression checks. No Blender or video-generation
  50. model is required.
  51. ```bash
  52. cp -r skills/hand-drawn-canvas-animation ~/.agents/skills/
  53. ```
  54. **[motion-reel](skills/motion-reel)** makes product films, showreels and UI morphs
  55. from a brief and visual references. It includes a Canvas renderer, spring motion,
  56. original music and SFX, beat analysis, contact sheets and MP4 checks. The skill
  57. covers planning, animation, review and export, with source files you can edit.
  58. More lands here as I clean it up.
  59. ---
  60. ## The rig everything is measured on
  61. 4x RTX 3090, 96 GB VRAM total, PCIe 3.0 x16, no NVLink, 320 W per card. Ubuntu.
  62. Numbers in these READMEs come from real runs on that machine, not from
  63. datasheets. Your absolute times will differ, the ratios usually hold.
  64. There is one thing worth checking on your own box before anything else. Cards
  65. ship at whatever power limit the vendor set, and it is often well under the
  66. default. Mine were at 220 W against a 350 W default, and raising them to 320
  67. took an identical run from 220 seconds down to 170.
  68. ```bash
  69. nvidia-smi --query-gpu=name,power.limit,power.default_limit --format=csv
  70. ```
  71. ---
  72. <div align="center">
  73. I post the benchmarks behind all of this. Local inference on 4x RTX 3090,
  74. open-weights models, argv and raw logs attached.
  75. **[@superalesha](https://x.com/superalesha)**
  76. MIT. Take it, change it, ship it.
  77. </div>

Ratings and comments

Sign in to rate this skill.

Sign in

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