| Filename | Latest commit message | Latest commit date |
|---|---|---|
| README.md | ||
Free AI Video Editing with Open Models (GLM + OpenCode)
Reference notes + link collection based on the YouTube video:
"My NEW AI Editor Completely FREE 0$ (No Subscription)" — Jad M.H | AI Automation (9:51) Video: https://www.youtube.com/watch?v=S5Y3Hp2P740
Note: the video's auto-captions render some names phonetically ("Claw" = Claude/Cl ... "Chad GBT" = ChatGPT, "GLM 5.3 Flash", "Opus 5.5"). Names below are kept as the creator states them.
Summary
The creator edited an entire video for literally $0 — no Claude/ChatGPT subscription, no API key, no credits, no account/card — using a free open model (GLM 5.3 Flash, stated as a 321B-parameter MIT-licensed model) inside OpenCode, the free open-source terminal coding agent. He claims to be the first on YouTube to show a full AI video edit done by a free open model.
Why this wasn't possible before: a real AI video edit needs an agent that can open files, run tools (ffmpeg etc.) on your machine, and stay on task for an hour — historically that meant Claude Code on a $20/mo plan plus credits, and free models were either too small to do a full edit or lost the thread on long jobs. Three things changed:
- OpenCode — free, open-source Claude-Code-style agent in the terminal. Crucially, it reads the same skill folders as Claude Code, so existing skills work without rebuilds.
- Free open models inside it — on Oct 16 OpenCode made GLM 5.3 Flash free to use on their servers (you don't need a beefy machine; the "brain" runs server-side).
- The skill (the real secret) — his "Editor V1" skill encodes his whole editing system: cut cadence, colors, sound, caption placement, and hard rules. The model doesn't need to be a genius; it just follows the recipe. Massive prompts are unnecessary ("complete BS") — you hand it the video and say which skill to use.
Transparency: all compute (cutting, transitions, graphics, rendering, sound mixing) ran on his MacBook; only the model ran on OpenCode's free servers. His edit used ~1.6M tokens at $0. Quality: "not as good as Opus 5.5, but it definitely does the work." He builds skills in Claude (smarter) and only runs them in OpenCode (cheaper). Of his editing skills, all work with the free model except the motion-graphics skill (every frame generated by code + music — too heavy for any open model today).
Install Steps (as demonstrated)
Works on Mac and Windows (OpenCode recommends WSL on Windows).
- Install OpenCode from the official install line at opencode.ai (terminal on Mac: Cmd+Space → terminal).
- Download the "Editor V1" skill zip (creator shares it free in his Skool community) — leave it in Downloads, no need to unzip.
- Open a fresh terminal, run
opencode. No sign-in needed — skip "connect" (that's for paid models). - Type
/models, search "free", pick GLM 5.3 Flash free. Press Ctrl+T and keep pressing until reasoning = max (editing is a big job). - Paste the setup prompt (from his Google Drive). Despite saying "Claude Code", OpenCode reads the same folder: it finds the zip, installs the skill, then installs the free tools — FFmpeg, Python, the video renderer, and a speech model for voice. It asks before anything needing a password, and asks if you want the ~1GB "cut-out effect" (person-in-front-of-background with a huge word behind the head). It finishes by rendering a 7-second test video as proof.
- Type
/newfor a fresh session so the skill reloads. - Drag in a video and prompt: "edit this video with editor v1". When done, reply "open it for me", watch, and give notes — the more notes, the better it gets.
Tips from the video
- Start short (his test was a 38-second clip) — don't feed it a 20-minute video first try.
- Give feedback notes with timestamps, and avoid decimal seconds ("at second 5", not "5.5").
- Before committing, ask "how long will it take?" — it answers up front.
Timestamps
| Time | Topic |
|---|---|
| 0:00 | The $0 challenge — free open model edited the whole video |
| 1:53 | Why it matters: paid tools ($20/mo + credits) block beginners |
| 2:24 | Why this wasn't possible a week ago (agent + long-task requirements) |
| 2:52 | The 3 things that changed: OpenCode, free GLM 5.3 Flash, the skill |
| 3:36 | The skill system — recipe/guardrails instead of massive prompts |
| 4:32 | Transparency: edit ran on his MacBook, model on OpenCode's servers |
| 5:17 | Which of his skills work (all but motion graphics); build skills in Claude, run in OpenCode |
| 6:27 | Step-by-step install (OpenCode, /models, Ctrl+T max reasoning, setup prompt) |
| 8:20 | Running the edit: "edit this video with editor v1" |
| 8:44 | Tips: start short, timestamped notes, ask for ETA first |
Link Collection
The video & creator
- Video: https://www.youtube.com/watch?v=S5Y3Hp2P740 (jump to setup steps: https://www.youtube.com/watch?v=S5Y3Hp2P740&t=387)
- Channel — Jad M.H | AI Automation: https://www.youtube.com/@JadM.H
- Creator's free Skool community (source of the Editor V1 skill + setup prompt; he says links are in the video description): https://www.skool.com
Tools used in the video
- OpenCode (free, open-source terminal agent): https://opencode.ai
- Docs: https://opencode.ai/docs
- Models docs (
/models, free models): https://opencode.ai/docs/models - Skills docs (skill folders, Claude Code compatible): https://opencode.ai/docs/skills
- Source code: https://github.com/anomalyco/opencode
- GLM open models (publisher — Z.ai / zai-org, MIT licensed): https://huggingface.co/zai-org
- FFmpeg (installed by the setup prompt — cutting/rendering): https://ffmpeg.org
- Python (tooling the skill drives): https://www.python.org
- Hugging Face (where the model is published/downloaded): https://huggingface.co
Comparables mentioned
- Claude Code (the paid agent the skills were built in): https://docs.anthropic.com/en/docs/claude-code
- OpenAI ChatGPT (credit-based alternative he contrasts with): https://chatgpt.com
Notes generated 2026-10-28 from the video transcript. Model/version names follow the creator's narration and may differ from official release names.