Pick your engines.
Wire them in minutes.

Native image + video inside the Grok harness — already in your SuperGrok plan.

Best for

• —Cinematic + photographic heroes generated from inside a Grok session
• —High volume on a flat monthly fee (no per-image bill)
• —Fast iteration when you are already working in Grok

Not when

Text-heavy/technical diagrams where labels must be legible — Grok Imagine fabricates text. Use Nano Banana Pro (Search-grounded) for those.

Auth

Logged-in Grok CLI/app. NB: the CLI may sit on a base-tier token showing "free credits %" — re-login with the main SuperGrok account for full quota.

Cost

✓ no extra API costIncluded in SuperGrok — no extra API cost

Included in

SuperGrok ($30/mo) — ~30–100 quality images/day on rolling windows

Rolling windows throttle bursts (50 then wait 2–6h). Video costs ~5–10× an image in quota.

When paid is worth it: For programmatic batches beyond the daily quota, the Grok Imagine API (~$0.02/image) supplements it.

Gemini 3 Pro Image built into Antigravity — free, in-editor, and Search-grounded.

Best for

• —Text-heavy + technical heroes, infographics, diagrams — Search grounding keeps labels legible & accurate
• —Real-time-data visuals (current pricing, dates, charts) via Google Search
• —No-extra-cost generation from inside Antigravity / a Gemini session

Not when

You are working in a non-Gemini harness with no key — Claude reaches the same model via the infogenius script path instead.

Auth

Logged-in Antigravity / Gemini account (no separate API key inside Antigravity)

Cost

✓ no extra API costFree in the Antigravity preview / included in Google AI subscription

Included in

Antigravity free public preview (NB2 free to all); Google AI Pro/Ultra for higher limits

When paid is worth it: Programmatic batches outside Antigravity → the Gemini API (per-image pricing) via the infogenius script.

Gemini-3 premium hero stills with Guardian AI routing — the in-house engine.

Best for

• —Hero imagery at the top of a page
• —Book + album covers that live for years
• —Research-grounded infographics (the InfoGenius pipeline)

Not when

Bulk B-roll or 20-attempts-in-10-minutes ad variants — that volume burns time at hero quality.

Auth

Google Generative AI API key (GOOGLE_GENERATIVE_AI_API_KEY)

Cost

Gemini API pricing (~pennies per image)

The durable path (scripts/lib/nb-image.mjs) needs no server and finds the key automatically.

One OAuth endpoint fronting 30+ frontier models — Sora, Veo, Kling, Soul.

Best for

• —Interactive cinematic video from inside the harness
• —Character-consistent B-roll via Soul ID training
• —Reaching Sora / Veo / Kling without juggling three API keys

Not when

High-volume programmatic batch — the subscription-credit model gets expensive, and MCP calls burn credits even in "unlimited" mode.

Auth

OAuth through your Higgsfield account (no API key to manage)

Cost

$15 Starter · $39–49 Plus · $99–129 Ultra · $89/seat Business

MCP generations always consume credits, even when the web UI shows "unlimited".

Pay-per-use aggregator — the economical lane for batch video and bulk B-roll.

Best for

• —Lyric-video batches across the 12k catalog
• —Bulk B-roll where per-generation cost matters
• —Programmatic generation from scripts (one API, many models)

Not when

A single premium hero still — use InfoGenius. Interactive cinematic exploration — use Higgsfield.

Auth

fal.ai API key (FAL_KEY)

Cost

30–80% cheaper than subscription credits on video; ~$0.025–0.05/image

OpenAI image gen built into Codex CLI via the $imagegen skill — on your ChatGPT plan.

Best for

• —Quick visuals while already coding in Codex
• —Strong text-in-image rendering on a sub you already pay for

Not when

Large batches — image turns burn the Codex limit ~3–5× faster. Switch to a per-call API for volume.

Auth

Codex signed in with a ChatGPT plan. Setting OPENAI_API_KEY switches it to paid API pricing.

Cost

✓ no extra API costIncluded in the ChatGPT plan (no extra cost); ~3–5× faster limit burn per image turn

Included in

ChatGPT (Free/Go/Plus/Pro/Business) — counts toward Codex usage limits

When paid is worth it: Batches of >250 images, or to avoid eating Codex coding quota → set OPENAI_API_KEY for gpt-image-2 API pricing.

OpenAI image lane — a second opinion via the key you already have.

Best for

• —A different aesthetic read when InfoGenius output feels off-lane
• —Strong text-in-image rendering
• —Teams already standardized on OpenRouter

Not when

Your default hero work — InfoGenius is the in-house premium tier and stays the default.

Auth

OpenRouter key (OPENROUTER_API_KEY — already on this machine)

Cost

OpenRouter per-image pricing

The raw frontier video models, direct — cheapest per call, most plumbing.

Best for

• —Squeezing per-call cost on one specific model at high volume
• —A production pipeline already standardized on one provider

Not when

You want one connector + Soul character consistency — Higgsfield fronts all three for far less setup.

Auth

Per-provider API keys (OpenAI / Google / Kling)

Cost

Lowest per-generation, but three separate integrations + no Soul consistency