Skip to content

106 social media skills for AI agents - strategy, writing, video, design, platform growth, publishing, and analytics. Works with Claude, Cursor, OpenClaw, Hermes & 40+ agents.

4.1 KB
Raw Blame

Luma Dream Machine — integration & connection guide

What it is: the cinematic AI video generator (Luma Labs — Ray3 model line) this stack uses for shot generation, image-to-video, and footage restyling. Luma generates shots; the human judges; capcut assembles; WoopSocial publishes. Skill: skills/luma/; the model-agnostic router is skills/ai-video/. Models, credit rates, and tier structures move fast and sources genuinely conflict — verify-quarterly at lumalabs.ai/pricing and the in-app plan pages.

Connection layers

  1. Dream Machine API + official SDK (lumaai-python — Apache-2.0, Python 3.9+, sync/async): create a generation → receive an ID → poll status → download. The agent can drive this where an API key exists. The wallet trap: API credits are billed separately and do NOT transfer from the app subscription — fund each independently. No unreviewed auto-publish regardless of automation.
  2. Web app + iOS (human executes): no key? The agent supplies shot briefs, reference plans, keyframe strategy, and the draft-first credit plan; the human generates in-app.
  3. Publish handoff: exports → assembly in capcut (pace, captions, licensed sound — Luma output is silent) → scheduling-and-queue → WoopSocial: upload media → attach → validate → create post (endpoints/tool names live in tools/integrations/woopsocial.md — don't hardcode).

The model line (≈, verify — it moves)

Ray3 = reasoning model, 16-bit HDR + EXR, keyframes, character reference, visual annotation, Modify (V2V) · Ray3.14 (Jan 2026) = the volume default — native 1080p, ~4× faster, ~3× cheaper at 720p, but no character reference or HDR (fall back to Ray3) · a newer Ray already appears on the official pricing page · Photon = Luma's image model · Luma Agents = the multi-model bundle (Veo 3.1, Kling, image + ElevenLabs audio models; brief-to-video from DOCX/PDF) — bundle math: worth it at 2–3 models used, overpaying at one.

Economics (the workflow IS the cost control)

Draft Mode → Hi-Fi master the winner — never iterate at delivery quality. Multipliers (≈, attribute): HDR 2×/HDR+EXR 3× the SDR rate; 1080p HDR ≈ 16× standard 720p; resolution jumps ≈ 4×. Budget 3 attempts per usable clip (4–5 for consistency shots). Monthly credits expire at reset (plan sessions); top-ups roll over (12-month validity reported); Trustpilot skews negative on billing — watch statements.

Rights + tiers (the reason this file exists)

Free = watermarked + explicitly NON-commercial. Whether the cheapest paid tier includes commercial rights conflicts across sources; the uncontested commercial floor is the Plus-level plan (~$30/mo, annual ~20% off) — and two plan generations coexist (legacy Dream Machine tiers still billable + 2026 Agents plans ~$30/$90/$300): verify in-app, never quote a tier as fact. The overlooked clause: you own outputs, but Luma retains a broad license to use your generations for service improvement and marketing — disclose to clients where relevant; Enterprise adds a data-privacy guarantee (not used for training) — the NDA tier. High-stakes → counsel (not legal advice).

Known limits + routing

No native audio (sound in post — ai-music-and-sound/licensed; sound-native generation → veo-3/kling) · 5–10s shots (Extend degrades past the initial clip — sequences assemble in capcut) · same-prompt variance >50% of runs (independent testing) — consistency via references, not re-rolls · avatar/presenter work → heygen/ synthesia · edit-grade control/editing → runway · brand-true stills in → flux/Photon.

Hard lines

Human judges every clip · no unpermitted real-person likeness (real or AI lookalike) · no photoreal fake-event footage targeting real people/companies · AI-disclosure where required (EU AI Act; C2PA) · commercial tier verified before client work; Luma's license-back disclosed where relevant · licensed sound only · never state a rate/tier as immutable fact · WoopSocial does not generate or edit video.