ai-cli — image + video gen via CLI

**Status:** Assimilation research note (no code change beyond the atlas record).

**Date:** 2026-06-23.

**Tagline (requester):** "Image + video gen via CLI."

**Method:** vendor homepage <https://ai-cli.dev> + npm/GitHub corroboration. Behavioral claims are vendor-stated and not independently exercised.

**Provenance (verified 2026-06-23):** open source, **Apache-2.0**, by **Vercel Labs** — github.com/vercel-labs/ai-cli, npm ai-cli (v0.3.1; npm self-describes as "agent-native" with "predictable artifact outputs").

**Atlas record:** tool:ai-cli — packages/atlas/graph/domain/tools/ai-cli.yaml (evidence evidence:ai-cli-homepage-2026-06, evidence:ai-cli-npm-2026-06).

---

0. Bottom line

A terminal CLI for generating **text, images, and video** from prompts, routed through the **Vercel AI Gateway** to models from OpenAI, Anthropic, Google, Black Forest Labs, and ByteDance. Built for piping and automation: multi-model comparison, stdin/stdout chaining, inline terminal previews, JSON metadata.

1. What it does

ai image/video/text "prompt" → generated output, with live model discovery and the ability to run the same prompt across multiple providers for comparison.

2. How it works

• Routes to providers via the **Vercel AI Gateway**; short model names auto-resolve (e.g. flux-2-pro, gpt-image-2).
• **Live model discovery** — no hardcoded model list (ai models).
• **Multi-model comparison** — same prompt across providers simultaneously.
• **Pipeable**: stdin/stdout chaining (e.g. ai image | ai video), inline terminal image preview via graphics protocols, JSON metadata output for automation.
• Minimal setup; configured via environment variables.

3. Install & usage

npm install -g ai-cli
ai image "prompt"
ai video "prompt"
ai text  "prompt"
ai models            # list available models

4. Relevance to babysitter / agent workflows

• A scriptable media-generation primitive an orchestrated agent could call for asset generation steps (e.g. producing diagrams/thumbnails/video in a pipeline), with JSON output suited to automation.
• The multi-provider gateway model overlaps conceptually with the repo's own provider/transport-adapter layer — useful as a reference point, not a dependency.
• Recorded as catalog awareness (used_for: skill-area:video-processing); not wired into babysitter.

5. Comparison (atlas edges)

No direct alternative_to recorded — no equivalently-scoped multi-provider media-gen CLI currently in the catalog. Conceptually adjacent to the LLM-gateway tools (tool:openrouter, tool:litellm) but those are text/inference proxies, not media-gen CLIs.

6. Caveats / open questions

• License/OSS/repo **verified** (Apache-2.0, Vercel Labs) via npm + GitHub. Early version (0.3.x).
• Provider/model list, pricing, auth model (env vars → which keys / Vercel AI Gateway billing), and video-gen latency/quality are vendor claims, not independently exercised here.