Sitelet https://slopcamera.com/docs/tutorials/codex
SlopCamera
Browse documentationTutorials · Set up SlopCamera for Codex

Tutorials

Set up SlopCamera for Codex

Install the CLI and Agent Skill, then ask Codex for a first diagram and a short video in a repository.

Codex can create diagrams, images, animations, and video through SlopCamera once two pieces are installed: the slopcamera command, which does the local media work, and the SlopCamera Agent Skill, which teaches Codex which operation to reach for.

You need Bun 1.3.14 or newer on macOS, Linux, or Windows. A few features are narrower: vectorization, local audio and color effects (media audio and media color), and the native engines (Blender, CadQuery, and Manim) run on macOS and Linux, and the GPU scene profiles need macOS. SlopCamera edits recordings you already have; it does not record. There is no SlopCamera account.

Install the CLI

bun add --global https://github.com/hraness/slopcamera/releases/download/v3.10.3/hraness-slopcamera-3.10.3.tgz

Confirm the install:

slopcamera doctor --json

The doctor reports which parts of SlopCamera work on this machine. Diagrams need nothing more, video editing needs FFmpeg and FFprobe, and native engines such as Blender install separately.

Install the Agent Skill for Codex

slopcamera skill install --target codex

Codex is the default target, so slopcamera skill install without --target does the same thing. A user-scope install writes ~/.codex/skills/slopcamera, which every Codex session can read. To keep the skill inside one repository, run this at its root instead:

slopcamera skill install --target codex --scope project

That writes .codex/skills/slopcamera in the repository. Pass --project <directory> to name a different root without changing directories, or --force to replace an existing install. The Codex target ships the same SKILL.md and task references as every other target, plus an agents/openai.yaml interface file whose default prompt invokes $slopcamera.

Start a new Codex session after installing. The skill loads version-matched guidance for choosing creative operations, so Codex knows to check a diagram before rendering it, to preview a cut before exporting it, and to keep sources editable.

Try it

Ask Codex in a fresh session:

Create a two-box diagram showing ingestion flowing into review, render it, and show me the dark PNG.

Codex should run slopcamera diagram init, edit the JSON source, run slopcamera diagram check --strict, then slopcamera diagram render, and open or report the resulting files. If it improvises with another tool instead, the skill did not load; start a new session or re-run the install.

Then ask for a short video:

Make an eight-second 1280 × 720 title animation that says "Launch week", render it to MP4, and show me the file.

With the skill loaded, Codex can write an HTML scene and a render request, check the request with slopcamera html render --dry-run --json, then render it. This needs the browser and FFmpeg runtimes that slopcamera doctor --json reports. Create and revise your first animation walks through the same steps.

What Codex can do next

  • Generate images, video clips, and narration through your own Vercel AI Gateway credential; see generate media.
  • Edit footage and deliver multiple aspect ratios; see edit video.
  • Direct spatial scenes and native films with the released CLI; see spatial scenes and native films.
  • Give Codex's MCP support fixed tools for diagrams, images, and scene inspection and planning; see MCP clients.

Troubleshooting

  • slopcamera: command not found: the global Bun bin directory is not on your PATH, or the install ran in a different shell. Re-open the terminal and re-run slopcamera doctor --json.
  • The skill is not loading: skills install per target. The Codex form installs to a .codex/skills directory; --target agents is the generic convention and --target claude is Claude Code. Re-run the command above, not a different target, then start a new session.
  • Install reports a legacy diagram skill: remove or move the old diagram directory inside the target's skills root, then re-run the install. SlopCamera will not place both skills side by side.
  • Generation asks for a key: model-backed operations use caller-owned Vercel AI Gateway access. Set AI_GATEWAY_API_KEY, or with a linked Vercel project use vercel env run -- <command> so the credential is injected for one command without being written to the project. This website never receives it. For prompt-only images, slopcamera credits topup and slopcamera credits wait set up prepaid Hraness Credits instead; see generate media.