Ariel is a tool for translating video voice tracks. It provides an easy to use web interface for uploading videos, creating translations, and redubbing the video with translated audio.

Ariel uses Google Gemini Text-To-Speech (TTS). This model is only available using the global Cloud Region. To simplify deployment, as certain versions of Gemini models are not available in every region, we also use the global region when sending data to Gemini. If you have strict policies on data residency, please be aware that the solution will require bespoke changes to meet your requirements and cannot be used as is.

The solution requires the following Google APIs and services to be enabled in your Google Cloud project:

• Vertex AI API: For advanced AI capabilities, including model hosting and management.
• Cloud Storage API: For scalable and secure object storage for videos and audio files.
• Cloud Logging API: For centralized logging, monitoring, and analysis of application logs.
• Cloud Text-to-Speech API: For converting text into natural-sounding speech.
• Cloud Translation API: For translating text between languages.
• Cloud Build API: For building and deploying the application container.
• Cloud Run API: For running the application in a serverless environment.
• Identity-Aware Proxy (IAP): For securing access to the application.
• IAM Service Account Credentials API: For managing service account credentials.
• Cloud Resource Manager API: For managing Google Cloud projects and resources.

Please ensure any organizational policies applied to your Google Cloud projects allow for all of these APIs and services before starting the deployment process.

Deploying Ariel involves two main steps: running the setup script to configure your Google Cloud environment and then running the deployment script to deploy the service to Cloud Run.

Before you begin, ensure you have the following installed and configured:

• Google Cloud SDK (which includes gcloud and gsutil)
• Authenticated with Google Cloud: gcloud auth login
• A Google Cloud project created and the billing enabled.

The setup.sh script is designed to automate the configuration of your Google Cloud project. It will:

• Prompt you for the GCP region to deploy the service in.
• Confirm your Google Cloud Project ID.
• Prompt for a Cloud Storage bucket name and create it for you.
• Enable all the necessary Google Cloud APIs.
• Create a dedicated service account for the Cloud Run service.
• Grant the required IAM permissions to the service account and your user account.
• Generate a configuration.yaml file with your project settings.

To run the setup script, execute the following command in your terminal:

./setup.sh

Follow the on-screen prompts to complete the setup process.

Once your environment is set up, you can deploy the application to Cloud Run using the deploy.sh script.

This script will:

• Read the configuration from configuration.yaml.
• Build the container image using Google Cloud Build.
• Deploy the service to Cloud Run with IAP (Identity-Aware Proxy) enabled for security.

To deploy the application, run the following command:

./deploy.sh

The script will output the URL of the deployed service once it's finished.

By default, only your user account will have access to the deployed application (via IAP). To grant access to other users, run the following command for each user:

gcloud projects add-iam-policy-binding YOUR_PROJECT_ID \
    --member="user:USER_EMAIL" \
    --role="roles/iap.httpsResourceAccessor"

Replace YOUR_PROJECT_ID with your Google Cloud project ID and USER_EMAIL with the email address of the user you want to grant access to.

To update an existing deployment, run git pull from your clone of the git repository and then run the deploy.sh script. The settings created for the initial deployment will be reused. If you wish to change any of the settings before updating, run the setup.sh script first, as during the initial Deployment.

Ariel is initially deployed to use the most recent, generally available Gemini model as of the last update to the code (currently Gemini 2.5). If you want to update the model being used, you will need to update the configuration.yaml file and redeploy the solution.

To update the model, follow these steps:

1. Edit the file configuration.yaml, updating the lines starting with GEMINI_PRO_MODEL and GEMINI_FLAH_MODEL to use the model code for the model you want to use. You can find model codes in the Gemini documentation.
2. If needed, you can also update the models being used for text-to-speech by updating the lines starting with GEMINI_TTS.
3. Redeploy the solution by running deploy.sh.

configuration.yaml is created by setup.sh during the initial deployment. If you no longer have this file, you can manually create it using the template and filling in the appropriate values for your project.

The first step required to run Ariel locally is to install the dependencies. We suggest following Python best practices and using a virtual environment. The requirements.txt file can be used with pip, or you can use uv to create the venv and install the dependencies. Please see the uv documentation or the (pip documentation)[https://packaging.python.org/en/latest/tutorials/installing-packages/] for more information.

Next, running setup.sh will create the configuration.yaml file required to set the necessary environment variable. Once the file is created, you can also run the application locally via therun_locally.sh script.

After running run_locally.sh, the solution will start on localhost, and generally on port 8080. You can use the UI just as you would if it were deployed to Cloud Run. Keep in mind that much of the processing will then happen locally, so the performance of the app will be directly related to the computer you are using.

setup.sh will create the file configuration.yaml, which the solution uses to configure things like the GCP project to use when sending requests to Gemini, the GCS buckets to use, and the like, as well as apply the necessary roles to service accounts and related admin tasks. If you are comfortable with Google Cloud, can set up the API and Storage bucket yourself, and do not want to have Service Accounts changed, you can also forego running setup.sh, make a copy of configuration.template.yaml named `configuration.yaml

Ariel is accessed using the web front end. To access it, navigate to the URL displayed at the end of the deployment process.

To dub a new video:

1. Click the "Select Video from Computer" button, select the video you want to dub from the file chooser, and click "Open" in your file chooser.
2. Add the speakers by clicking the "⊕ Add Speaker" button. You must add a speaker for each speaker in the video, in the order they start speaking.
   1. In the Select Speaker Voice dialog, you can perform several actions:
      • Add a label for the speaker in the first text box.
      • Search for a specific voice using the second text box.
      • Preview a voice by clicking the play button next to it.
      • Filter voices by gender by clicking the "Male" and "Female" buttons.
   2. To select a voice, click on it in the list and then click the "Add Voice" button.
   3. The selected voice will appear in the list of speakers. Repeat this for every speaker in the video.
3. Select the original language of the video.
4. Select the language you wish to translate the video to.
5. You can choose to use the pro or flash version of Gemini using the toggle labeled "Gemini Model".
6. You can choose to automatically speed up or slow down the translated speech to match the original by using the "Adjust Speed" toggle. Please note that this does not normally lead to the best results.
7. If you have any additional instructions for the initial translation or dubbing, enter them in the text box labeled "Gemini Instructions". This can include things like words that should not be translated or the general tone of voice to use.
8. Click the Start Processing button to start the initial dubbing. This can take a few minutes, so please be patient. If an error occurs, a message will be shown. Please try processing the video again and, if it continues to not work, open a bug so that we can help you.

Once the video has been processed, the edit screen will be displayed to allow you to edit the dubbing.

In the Video Settings section you can change the initial settings for the transcription, including the languages and voices.

On the timeline, you can drag utterances to change the starting time.

In the list of utterances, you can:

• remove an utterance from the final video using the trashcan icon.
• use the original audio in the final video via the mute icon.
• edit the utterance using the pencil icon

After clicking the pencil icon for a specific utterance, there are a number of actions you can take.

• Edit the original transcription by changing the text in the Original Text text area.
• Listen to the translated audio by clicking the speaker icon.
• Edit the translation by changing the text in the Translated Text text area.
• Update the translation by providing additional instructions in the Translation Instructions text area and then clicking the Regenerate Translation button.
• Update the dubbed audio by providing additional instructions in the Voice Intonation Instructions text area and clicking the Regenerate Dubbing button.
• Change the start time of the translated utterance in the final video by editing the number in the Translated Start Time text box.
• Change the end time of the translated utterance in the final video by editing the number in the Translated End Time text box.
• Change the speaker by selecting a new speaker in the Speaker drop down. You will also need to click the Regenerate Dubbing button to update the audio with the new speaker.

Once you are happy with your edits, you must click the Save button to save the changes and have them applied to the final video.

Regenerating the translation will also automatically regenerate the dubbing so that you can hear the updated audio before deciding to save it.

Changing the start or end time will automatically update the other time to ensure the overall duration of the clip is correct.

Once you are satisfied with the individual utterances, click the Generate Video button to create the final dubbed translation. This can take some time, to please be patient.

Once the video is complete, it will be shown on the screen. You can preview the video by clicking the play button in the video player.

To download the video or the audio, click the appropriate button on the screen.

If you aren't satisfied with the final output, you can click the Go back to editing button to make additional changes.

To start dubbing a new video, you can click the Start Over button to return to the app's start page.

The exact costs for Ariel will be directly proportional to the length of the video and the amount of spoken text it contains.

Our experience shows that a 10 second video with one speaker will use approximately:

• 1000 - 1500 tokens for transcription
• 600 - 1000 tokens for translation
• 100 - 200 characters for text-to-speech

In addition, there are costs related to Cloud Run each time a video is processed, Cloud Storage for storing the videos and intermediate files, and initial Cloud Build costs for deploying the solution. These costs will be dependent on the region and specifics of your Cloud project.

At the time of writing (December 2025), dubbing a 30 second video costs less than 0.50 USD.

For information on Gemini pricing, please see the official documentation.

For information on the costs of the other services used by Ariel:

• Google Cloud Storage pricing
• Google Cloud Run pricing
• Google Cloud Build pricing