# MoneyPrinterTurbo Integration with OfoxAI – Setup Guide

Source: https://ofox.io/docs/integrations/moneyprinterturbo

[MoneyPrinterTurbo](https://github.com/harry0703/MoneyPrinterTurbo?utm=ofox_ai) turns a topic or keyword into a finished short video: it writes a script, sources or generates matching footage, adds subtitles and a soundtrack, then renders the clip. OfoxAI is an official sponsor of the project and ships as a built-in option for two of its stages — AI-generated video material and script writing — both reachable with one OfoxAI API key.

This guide only covers pointing MoneyPrinterTurbo at OfoxAI. For installing the tool itself — via `uv`, Docker, the Windows one-click package or Google Colab — see the [official README](https://github.com/harry0703/MoneyPrinterTurbo?utm=ofox_ai).

## Prerequisites

- An OfoxAI account with an API Key ([Get one here](https://app.ofox.io))
- MoneyPrinterTurbo installed, with `config.toml` present in the project root — the app copies it from `config.example.toml` on first run

## The one-prompt agent path

MoneyPrinterTurbo publishes an official agent skill. Any AI agent that can run terminal commands — Claude Code among them — installs, configures and renders from a single prompt:

```text
Use this skill: https://raw.githubusercontent.com/harry0703/MoneyPrinterTurbo/main/docs/skill/SKILL.md
Make me a video about how AI is changing everyday life, and generate the footage with OfoxAI.
```

The agent reuses whatever is already in `config.toml` and asks only for the credentials it still needs — `MPT_OFOX_API_KEY` for OfoxAI.

Mind the `MPT_` prefix: `MPT_OFOX_API_KEY` belongs to the agent path only. It is **not** the `OFOX_API_KEY` that `config.toml` falls back to in Step 1 below — setting the wrong one leaves the agent asking for a key you think you already provided.

The skill defaults to free Pexels stock footage, so ask for OfoxAI footage explicitly as above; otherwise the clip is cut from stock video and no OfoxAI video call is made. It runs on macOS and Windows and uses `uv`.

## Configuration steps

### Step 1: Set OfoxAI as the video source

Open `config.toml` and point `video_source` at `ofox`:

```toml filename="config.toml"
video_source = "ofox"

ofox_api_key = "<your_OFOXAI_API_KEY>"
ofox_base_url = "https://api.ofox.io/v1"
ofox_text_to_video_model = "bytedance/seedance-2.0-fast"
ofox_resolution = "720p"
ofox_min_duration = 4
ofox_max_duration = 15
ofox_provider = "byteplus"
```

`ofox_resolution`, `ofox_min_duration`, and `ofox_max_duration` are only valid for the model set in `ofox_text_to_video_model` — every model has its own accepted range. Switching models without updating these three fields gets the request rejected before any task is created. Check each model's real limits in [Video models](https://ofox.io/docs/api/videos/models), and what `ofox_provider` accepts in [Provider routing](https://ofox.io/docs/api/videos/provider-routing).

Leave `ofox_api_key` empty and export `OFOX_API_KEY` in the environment instead if you'd rather not store the key in the file — MoneyPrinterTurbo falls back to that variable.

Or do it from the WebUI: pick **OFox AI Video** as the source on the main screen, then open **Settings → Material sources** and scroll down to the **OfoxAI** block:

![Settings → Material sources tab, scroll down to find the OfoxAI block](https://ofox.io/screenshots/moneyprinterturbo/01-material-source-tab.png)

Fill in the OFox API Key, OFox text-to-video model, OFox Base URL, and OFox upstream provider:

![The four fields in the OfoxAI block](https://ofox.io/screenshots/moneyprinterturbo/02-ofox-fields.png)

The WebUI exposes only four of these fields — `ofox_api_key`, `ofox_text_to_video_model`, `ofox_base_url` and `ofox_provider`. `ofox_resolution`, `ofox_min_duration` and `ofox_max_duration` have no UI control at all; set them in `config.toml`.

### Step 2: Set OfoxAI as the script-writing model (optional)

MoneyPrinterTurbo's built-in `openai_*` fields work with any OpenAI Chat Completions-compatible endpoint — including OfoxAI's, reusing the same API key:

```toml filename="config.toml"
llm_provider = "openai"

openai_api_key = "<your_OFOXAI_API_KEY>"
openai_base_url = "https://api.ofox.io/v1"
openai_model_name = "openai/gpt-5.5"
```

Unlike `ofox_api_key`, `openai_api_key` has no environment-variable fallback — paste the key directly into the file.

The same three values live under **Settings → Models** in the WebUI — click **Configure LLM** next to the video topic field on the main screen to open it:

![Clicking Configure LLM opens the settings dialog](https://ofox.io/screenshots/moneyprinterturbo/03-llm-provider-entry.png)

There's a **Test model connection** button on that page:

![Settings → Models tab](https://ofox.io/screenshots/moneyprinterturbo/04-llm-provider-settings.png)

Model IDs need the vendor prefix — `openai/gpt-5.5`, not the bare `gpt-5.5`. Browse the full catalog in the [Model Marketplace](https://ofox.io/models).

### Step 3: Start MoneyPrinterTurbo

Restart the app (or the container, under Docker) so it picks up the edited `config.toml`. Generate a short test video from the WebUI or CLI; a completed render confirms both stages are reaching OfoxAI.

Docker deployments read `config.toml` at container start, not from a live-reloaded mount — edit the file and copy it into place *before* `docker compose up`, then restart the container after any later change.

## What it costs

OfoxAI footage is billed **per clip submitted, not per finished video** — and one video is normally cut from several clips, so a single render can create several paid tasks. Budget against the clip count, not the video count.

Both paths gate the spend. The WebUI estimates how many paid tasks your script will submit, then blocks rendering until you tick **I understand that OFox video generation creates paid tasks**. The agent skill halts on `OFOX_CHARGE_CONFIRMATION_REQUIRED` and only proceeds once you approve, by rerunning with `--confirm-ofox-charge`; a well-behaved agent never adds that flag on its own.

Per-second rates differ by model, from `bytedance/seedance-2.0-mini` upward; see [Video models](https://ofox.io/docs/api/videos/models). Script writing is billed as ordinary chat tokens, which is negligible next to the footage.

## Recommended models

- **Video** — any of the [12 OfoxAI video models](https://ofox.io/docs/api/videos/models) works in `ofox_text_to_video_model`; `bytedance/seedance-2.0-fast` (the default above) is the cheapest starting point, and `alibaba/wan-3.0` or `bytedance/seedance-2.5` extend to 30-second clips.
- **Script writing** — any chat model in the [Model Marketplace](https://ofox.io/models) works in `openai_model_name`.

## Troubleshooting

**Task creation fails immediately, before any billing** — `ofox_resolution`, `ofox_min_duration`/`ofox_max_duration`, or `ofox_provider` don't match the model set in `ofox_text_to_video_model`. The API validates these against the model's own limits and rejects a mismatch with `provider_type_unavailable` or a resolution/duration error before creating a task — nothing is billed. Check the model's real limits in [Video models](https://ofox.io/docs/api/videos/models).

**The agent keeps asking for a key you already set** — you set `OFOX_API_KEY` where the agent path wants `MPT_OFOX_API_KEY`, or the reverse. The two are separate: `MPT_OFOX_API_KEY` for the skill, `OFOX_API_KEY` as the `config.toml` fallback.

**Connection failing in mainland China** — switch `ofox_base_url` to the mirror `https://api.ofox.io/v1`; it mirrors the same service under the same API key.

**Script generation silently uses the wrong provider** — `llm_provider` must be exactly `"openai"` to activate the `openai_*` block; any other value routes script generation to a different provider's fields instead, even with `openai_api_key` filled in.

MoneyPrinterTurbo iterates quickly and its configuration fields may change between versions. Cross-check against the [official config.example.toml](https://github.com/harry0703/MoneyPrinterTurbo/blob/main/config.example.toml?utm=ofox_ai).