SoeMind Forge — AI Content Studio for Claude Code, Codex, Gemini CLI, Hermes & OpenClaw

Generate AI videos (Veo 3.1), images (Nano Banana), voiceovers, music, and
30-day content calendars from plain English — budget-aware: it asks before it
spends, previews before it renders, and keeps your brand consistent down to
the character's face.

Describe what you want in plain English — "create a TikTok ad for my coffee
brand", "make a cinematic story film" — and your AI coding agent plans,
prices, generates, QAs, and packages it ready to upload.

Cinematic Films • Kinetic Text Reels • Videos • Carousels • Voiceover • Music • 30-Day Calendars

Why this beats raw prompting an AI agent:

• Money moves only with consent. Every project has a budget cap and a spend
  ledger; the agent shows the estimated cost before generating, previews cheap
  before rendering expensive, and records every dollar.
• Consistency is engineered, not hoped for. Characters, products, logos, and
  voices are generated once, locked in an asset registry, and reused by ID — the
  same face across a whole campaign, not a new stranger per clip.
• Text never garbles. AI models paint beautiful backgrounds but butcher
  typography — so slides and reels get pixel-perfect text rendered locally with
  real fonts (Remotion), for $0.
• It ships upload-ready. Correctly sized media per platform, captions within
  char limits, hashtags, alt text — not a folder of loose renders.
• Proof stays real. The agent is hard-blocked from inventing testimonials,
  stats, or press. Your numbers, verbatim, or nothing.

Who this is for: founders, marketers, and small brands who want professional
AI content without babysitting the spend — no coding required.

What you need:

1. A free Google AI Studio API key —
   this kit is optimized for Gemini models. The free tier works to start;
   pay-as-you-go unlocks higher-quality video and images. Everything core
   (video, images, voiceover, music, transcription) runs on Gemini alone.
   Optional add-ons: OpenRouter — more models
   (GPT-4, Claude) plus Seedance 2.0, which makes a talking avatar from a
   custom voice audio file you supply; and RunPod
   — InfiniteTalk talking avatar from a custom/own voice with no duration
   cap. The agent wires either in on request.
2. Any supported AI agent CLI — Claude Code, Codex CLI, Gemini CLI,
   Hermes, or OpenClaw
3. Node.js 18+

What it costs: you pay only the AI providers, per generation. Rough guide: a
thumbnail ≈ $0.07, a 30-second voiceover ≈ $0.01, an 8-second Veo 3.1 clip ≈
$0.80, a full 32-second cinematic brand film (character sheet, keyframes, 4
clips, adaptive narration) ≈ $3.65. Every project gets a budget cap + ledger so
the agent warns before big generations. See Cost Reference.

Supported Providers:

• Google AI Studio (Gemini) — the optimized core. Veo video, Gemini 3 image
  models (Nano Banana), Lyria music, TTS, transcription, and Omni Flash speaking
  characters (talking avatar using Gemini's own voice) all run here. A free
  key gets you started; pay-as-you-go is only for higher-quality video/image
  generations. You can run the whole kit on Gemini alone.
• OpenRouter (optional) — a gateway to more models: 100+ text models (GPT-4,
  Claude, Llama) plus Seedance 2.0, which drives a talking avatar from a
  custom voice audio file you supply. The kit is tuned around Gemini, so the
  agent adds these on request.
• RunPod (optional) — InfiniteTalk: a talking avatar from a custom
  voice, even your own recording (same idea as Seedance) with no duration
  cap — though keep clips ≤60s, since longer runs raise the odds of model or
  video glitches. Sign up (referral link: load
  $10+ and get a $5–500 one-time bonus credit — most get $5–10; supports this
  project), then create a key under
  Settings → API Keys

New here? Follow GETTING-STARTED.md — from zero to
your first generated image in about 10 minutes.

Quick Start

macOS / Linux

./setup.sh

Windows

setup.bat

Manual Setup

# 1. Install dependencies
npm install
cd gemini && npm install && cd ..
cd workflows && npm install && npm run build && cd ..

# 2. Set API keys
cp .env.example .env
# Edit .env and add:
#   GEMINI_API_KEY=...      (Google AI Studio)
#   OPENROUTER_API_KEY=...  (OpenRouter - optional)

# 3. Start your AI agent (setup.sh installed the skills for your tool)
claude   # loads CLAUDE.md context automatically
> Create a TikTok ad

# other tools (hermes, codex, gemini):
> Read AGENT-GUIDE.md first, then create a TikTok ad

What You Can Create

Provider Comparison

Gemini is the optimized core — the whole kit runs on it. OpenRouter and
RunPod are optional add-ons.

When to use which:

• Gemini (default) — video, images, music, TTS, transcription. Start and
  finish here; it's what everything is tuned for.
• OpenRouter (optional) — models Gemini doesn't offer (GPT-4/Claude text) and
  Seedance 2.0, which lip-syncs a talking avatar to a custom voice audio
  file you provide (clip-length limited, ~4–15s). Agent-added on request.
• RunPod (optional) — InfiniteTalk: talking avatar from a custom/own
  voice with no duration cap (keep ≤60s for reliability). Pick this over
  Seedance for long-form or your own recorded voice.
• Talking avatar — whose voice? Gemini (Omni Flash) = Gemini's own voice
  only; Seedance & InfiniteTalk = your custom voice. Non-English/Myanmar
  pronunciation → Gemini.
• Veo 3.1 vs Omni Flash - Veo for cinematic no-dialogue/English beats and
  first+last-frame shots; Omni Flash for non-English speaking characters
  (e.g. Myanmar — better pronunciation, no silent filter blocks), reference
  consistency, and editing existing clips. Details:
  workflows/VIDEO-PROMPT-GUIDE.md § Production-Tested Playbook.

Platform Specs (All Supported)

See workflows/PLATFORM-SPECS.md for complete specs.

Skills Available

Documentation

Prompt Guides

Folder Structure

project-root/
├── .env                  # API keys (gitignored)
├── AGENT-GUIDE.md             # AI agent entry point
├── USER_GUIDE.md         # User documentation
├── setup.sh / setup.bat  # Setup scripts
│
├── workflows/            # Workflow functions + guides
│   ├── PLATFORM-SPECS.md     # All platform sizes
│   ├── PROMPT-GUIDES-INDEX.md # Master guide index
│   ├── IMAGE-PROMPT-GUIDE.md
│   ├── PRODUCT-SHOT-GUIDE.md
│   ├── VIDEO-PROMPT-GUIDE.md
│   ├── THUMBNAIL-GUIDE.md
│   ├── BRAND-ASSETS-GUIDE.md
│   └── STYLE-GUIDE.md
│
├── skills/               # Generation skills
│   ├── generate-video/
│   ├── generate-image/
│   ├── generate-brand-assets/
│   ├── generate-voiceover/
│   ├── generate-music/
│   ├── content-preflight/
│   ├── content-review/
│   ├── plan-content/
│   └── write-copy/
│
├── templates/            # Project templates
│   ├── ASSETS-GUIDE.md       # What to store
│   ├── assets.config.template.yaml
│   ├── project.md
│   └── brand.md
│
├── projects/             # User projects (gitignored)
│   └── {project-name}/
│       ├── templates/    # Filled project info
│       ├── config/       # Asset registry
│       ├── assets/       # Reference files
│       └── output-contents/
│
├── gemini/               # Gemini API provider
└── openrouter/           # OpenRouter provider

Supported AI Tools

One unified layer: skills — the same skills/*/SKILL.md files drive every
supported tool.

Run setup once and pick your tool — it installs the skills to the right place.
Claude Code additionally loads CLAUDE.md automatically for
context; on other tools, tell the agent to read AGENT-GUIDE.md first.

For AI Agents

Claude Code reads CLAUDE.md automatically — just ask:

Help me create a video ad.

Other tools — always tell the agent to read first:

Read AGENT-GUIDE.md first, then help me create a video ad.

Read AGENT-GUIDE.md and workflows/PROMPT-GUIDES-INDEX.md, then create a 30-day content calendar.

Asset System

Store reusable assets for consistency across content:

See templates/ASSETS-GUIDE.md for complete guide.

Cost Reference

License

MIT © Min Khant Soe — Senior AI Developer,
Thailand. Founder, SoeMindAI, Inc.

Contributions welcome — see CONTRIBUTING.md.