krusemediallc/arcads-claude-code

What this repo does

Arcads AI Video – Agent Skill Pack is a collection of ready‑to‑run Claude Code (and Cursor) “skills” that let you generate advertising‑focused images and short videos with the Arcads generative‑media platform. It wires together:

• Video models – Seedance 2.0, Sora 2, Veo 3.1, Kling 3.0, Grok Video, OmniHuman, Audio‑driven – each exposed through a small CLI‑style skill that handles prompting, API calls, polling, cost confirmation and file‑saving.
• Image models – Nano Banana 2/Pro/Edit and ChatGPT Image 2 – used for influencer character sheets, product stills, and the 37‑template library of static Meta ads.
• Multi‑step pipelines – e.g., storyboard generation → Seedance video → ffmpeg stitching → caption burn‑in, or Pixar‑style 3‑D‑look ads, clay‑mation style ads, YouTube thumbnail batches, etc.
• Meta ad publishing – a meta‑ad‑builder skill that uploads the finished creative to the Meta Marketing API as a paused ad.

All of this is packaged as a set of Claude Code skills (also mirrored for Cursor) plus a handful of Bash/Node helper scripts. The repository does not include the Arcads service itself; you need an Arcads account and API key, which the one‑time setup.sh script stores in a local .env file.

---

Who might use it

• Media buyers / performance marketers who want to automate the creative production of Facebook/Meta ads.
• Freelance ad creators looking for a repeatable workflow to generate influencer‑style video, UGC‑style clips, or high‑impact static image ads.
• Developers who want a concrete example of how to orchestrate multiple generative‑AI models (video, image, audio) in a single end‑to‑end pipeline.

---

How to get started (quick‑start steps)

1. Clone & run the installer

   git clone https://github.com/krusemediallc/arcads-claude-code.git
   cd arcads-claude-code
   ./scripts/setup.sh
   

   The script prompts for your Arcads API key, writes it to .env, verifies connectivity, and creates a personal MASTER_CONTEXT.md workspace file.
2. Open the folder in Claude Code (or Cursor) – the editor runs a SessionStart hook that shows which skills are available and whether your environment is ready.
3. Pick a workflow – the README lists ready‑made prompts, e.g.:
   • Seedance 2.0 UGC selfie – a 12‑second product‑review video.
   • generate‑youtube‑thumbnail – five high‑CTR thumbnail formulas.
   • pixar‑style‑ad – storyboard → Seedance video → ffmpeg stitching.
4. Run the skill – either via Claude Code’s UI or by invoking the corresponding Bash/Node script. The agent will:
   • ask for any missing parameters (product name, reference images, etc.),
   • show an estimated credit cost, and
   • launch the Arcads API call, poll for completion, and write the output files to the outputs/ folder.
5. (Optional) Publish – use the meta‑ad‑builder skill with your Meta access token to upload the creative as a paused ad.

---

Key technical details

Area	What the repo provides
Core language	Pure Python 3.10+ for the Arcads API wrapper; most image‑ad scripts are std‑lib only (no external packages).
CLI helpers	Bash scripts (setup.sh, check‑arcads‑env.sh, sync‑skill.sh) and a few Node‑based steps (npx hyperframes for caption burn‑in).
External tools	ffmpeg (video stitching, chroma‑key caption overlay), jq (JSON handling), whisper (audio transcription), Node.js (HyperFrames).
Skill organization	Each capability lives under skills/… (e.g., skills/arcads-external-api/, skills/generate-youtube-thumbnail/). Shared utilities and prompt libraries are in shared/skills/.
Prompt libraries	Ready‑made markdown prompt templates for each video model (Seedance, Sora, Veo, etc.) and for the 37 static‑ad templates.
Cost tracking	Every API call is logged to logs/arcads‑api.jsonl with creditsCharged; the agent reads this to give accurate cost estimates before generation.
Reference assets	references/ (git‑ignored) holds influencer photos, product shots, and style reference folders that the agent automatically up‑scales if needed.
Stateful workspace	MASTER_CONTEXT.md acts as a persistent memory file – the agent reads/writes product defaults, brand voice, credit totals, and session notes.

---

When it’s useful vs. when you might look elsewhere

Use this repo if you already have an Arcads account and want a plug‑and‑play way to drive its video/image models from a local development environment, especially for ad‑creative production.

Skip it if you:

• Need a generic, model‑agnostic generative‑AI framework (this repo is tightly coupled to Arcads).
• Want a fully hosted SaaS solution without any local setup (you’d use Arcads’ own UI instead).
• Are looking for a pure‑Python library without the Claude Code skill scaffolding.

---

Quick reference of the main models supported

Model	Type	Typical use‑case
Seedance 2.0	Video (4‑15 s)	Flagship ad clips – UGC selfie, premium product reveal, hero shots, lookbooks, feature demos.
Sora 2	Video (up to 20 s)	Longer text‑to‑video narratives; can remix existing assets.
Veo 3.1	Video (~8 s)	Animate a single‑frame influencer still into a short clip.
Kling 3.0	Video (5‑10 s)	B‑roll or generic scene generation.
Grok Video	Video	xAI’s text‑to‑video model.
Nano Banana 2 / Pro	Image	Influencer character sheets, product stills, static ad creatives.
ChatGPT Image 2	Image	UI‑style, typography‑heavy static Meta ads.
OmniHuman / Audio‑driven	Video	Talking‑avatar or lip‑sync videos.

---

Bottom line

arcads-claude-code is a real, functional toolkit that stitches together a suite of generative‑AI models (video, image, audio) into production‑ready advertising pipelines, all orchestrated through Claude Code/Cursor skills. If you’re a marketer or developer comfortable with a little terminal work and want to automate the creative side of Meta ad campaigns, this repo gives you a complete, documented starting point.