Public Summary of Training Content: Ray Models

Last updated: October 23, 2026

Version of the Summary:
V. 1

Last update:
October 2026

1. General information

1.1. Provider identification

Provider name and contact details:
Luma AI, Inc., 601 Marshall Street, Redwood City, CA 94063, United States. Contact: legal@lumalabs.ai

Authorised representative name and contact details:
European Data Protection Office, Avenue Huart Hamoir 71, 1030 Brussels, Belgium. Contact: info@edpo.com

1.2. Model identification

Versioned model name(s):
This Summary covers the following models of Luma’s Ray model family (API identifiers in brackets): Ray 3 and Ray 3 HDR (ray-3, ray-hdr-3); Ray 3.14 and Ray 3.14 HDR (ray-3-14, ray-hdr-3-14); and Ray 3.2 (ray-3.2). Luma’s Uni models are covered by a separate Summary.

Model dependencies:
Ray 3.14 and Ray 3.2 are derived from Ray 3. These models are covered by this Summary. Ray 3 was initialised from publicly released open-weight model weights made available by a third party under an open-source licence, and was subsequently further trained by Luma on the data described in this Summary. This Summary describes the data used by Luma in its further training of the Ray models. It does not describe the data used by the third party to train the original open-weight model.

Date of placement of the model on the Union market:
Ray 3: 4 September 2025; Ray 3.14: 22 January 2026; Ray 3.2: 26 May 2026.

1.3 Modalities, overall training data size and other characteristics

Modality Select the modalities present in the training data, to the extent that they are identifiable	Training data size For each selected modality, select the range within which the estimated total training data size for that modality falls. Dynamic datasets may be excluded from the estimation.	Types of content For each selected modality, provide a general description of the type of content that has been included in the training data.
☒ Text	☐ Less than 1 billion tokens ☒ 1 billion to 10 trillion tokens ☐ More than 10 trillion tokens Alternatively, specify the approximate size in a different measurement unit: Not applicable.	Text paired with images and video (e.g., captions, descriptions and instructions), including licensed curated text corpora paired with audiovisual material, publicly available text, and synthetic captions and instructions.
☒ Image	☐ Less than 1 million images ☐ 1 million to 1 billion images ☒ More than 1 billion images	Photography, visual art and design works, illustrations and stock imagery, including professionally produced licensed content and publicly available imagery, as well as synthetic images generated by Luma’s own models.
☐ Audio	☐ Less than 10 000 hours ☐ 10 000 to 1 million hours ☐ More than 1 million hours	Not applicable.
☒ Video	☐ Less than 10 000 hours ☐ 10 000 to 1 million hours ☒ More than 1 million hours	Film and cinematic footage, stock video, advertising and other creative content, and general-interest video clips (including cinematic aerial footage and street-level scenes), as well as synthetic video generated by Luma’s own models and scripted simulations.
☐ Other	Not applicable.	Not applicable.

Latest date of data acquisition/collection for model training:
For each model, no later than its date of placement on the Union market (see Section 1.2). The models are not continuously trained on new or dynamic data after that date.

Description of the linguistic characteristics of the overall training data:
Text in the training data is predominantly in English. Other languages, including EU official languages, are represented to a limited extent.

Other relevant characteristics of the overall training data:
The training data consists predominantly of natural, human-scale, visible-spectrum imagery and video of the kind used in creative, entertainment and advertising content.

Additional comments (optional):
Data sizes are estimates after pre-processing and include licensed, publicly available and synthetic data. Video duration was estimated from total clip length.

2. List of data sources

2.1. Publicly available datasets

This Section requires information about datasets that were used to train the model and which have been compiled by a third party, are made available publicly for free, and are readily downloadable as a whole or in predefined chunks, such as datasets and collections available on public repositories and online platforms, specialised websites, or snapshots of common crawl. The public availability of the datasets for free does not mean that the content at issue is necessarily free of rights since it may be subject to licensing arrangements or conditions of use (e.g., certain free and/or open licenses may determine the scope of the uses, including prohibiting uses relating to model training).

A dataset is considered to be “large” if the total data size for any one of the modalities contained in the dataset exceeds 3% of the size of all publicly available datasets for that modality used for training. The size of the dataset should be based on its size after pre-processing (for example filtering), and without splitting the dataset to prevent reporting circumvention.

Have you used publicly available datasets to train the model?
☒ Yes ☐ No

If yes, specify the modality(ies) of the content covered by the datasets concerned:
☒ Text ☒ Image ☒ Video ☐ Audio
☐ Other If so, please specify…

List of large publicly available datasets:
N/A

General description of other publicly available datasets not listed above:
Publicly available image and text datasets compiled by third parties, including collections of images paired with captions and general-purpose text corpora.

Additional comments (optional):
—

2.2 Private non-publicly available datasets obtained from third parties

2.2.1. Datasets commercially licensed by rightsholders or their representatives

Have you concluded transactional commercial licensing agreement(s) with rightsholder(s) or with their representatives?
☒ Yes ☐ No

If yes, specify the modality(ies) of the content covered by the datasets concerned:
☒ Text ☒ Image ☒ Video
☐ Audio ☐ Other If so, please specify…

2.2.2. Private datasets obtained from other third parties

Have you obtained private datasets from third parties that are not licensed as described in Section 2.2.1, such as data obtained from providers of private databases, or data intermediaries?
☐ Yes ☒ No

If yes, specify the modality(ies) of the content covered by the datasets concerned:
☐ Text ☐ Image ☐ Video ☐ Audio ☐ Other If so, please specify…

If publicly known, list private datasets obtained from other third parties:
Not applicable.

General description of non-publicly known private datasets obtained from third parties
Not applicable.

Additional comments (optional):
—

2.3 Data crawled and scraped from online sources

Were crawlers used by the provider or on behalf of?
☒ Yes ☐ No

If yes, specify crawler name(s)/identifier(s):
Luma’s crawler did not use a dedicated name. It used the default user-agent string of the Python Requests library, in the form “python-requests/[version number]”.

Purposes of the crawler(s):
Collection of publicly available images, video and text for the training of Luma’s generative AI models.

General description of crawler behaviour:
Limited to a list of specific websites selected by Luma, rather than a general crawl of the web. Each website was reviewed by a member of Luma’s staff before it was added to the list, and websites known for hosting pirated or otherwise copyright-infringing content were not added. Luma’s crawler did not circumvent CAPTCHAs, paywalls, or other technical access restrictions, and did not access private content that requires a login.

Period of data collection:
From October 2023 to April 2026. Luma is not currently crawling online sources to collect training data.

Comprehensive description of the type of content and online sources crawled:
Crawled content includes a broad range of publicly accessible online material, including reference, educational, scientific, technical, government and institutional, and general-interest content. Crawled sources include text, video, images, and associated metadata such as captions, alt text, transcripts, or other descriptive text, and were filtered for quality and safety before training.

Type of modality covered:
☒ Text ☒ Image ☒ Video ☐ Audio
☐ Other If so, please specify…

Summary of the most relevant domain names crawled:
The most relevant crawled source domains include image-rich sites, document and PDF repositories, and audiovisual media.

Additional comments (optional):
—

2.4 User data

Was data from user interactions with the AI model (e.g. user input and prompts) used to train the model?
☐ Yes ☒ No

Was data collected from user interactions with the provider’s other services or products used to train the model?
☐ Yes ☒ No

If yes, provide a general description of the provider’s services or products that were used to collect the user data:
Not applicable.

Type of modality covered:
☐ Text ☐ Image ☐ Video ☐ Audio
☐ Other If so, please specify…

Additional comments (optional):
—

2.5 Synthetic data

Was synthetic AI-generated data created by the provider or on their behalf to train the model?
☒ Yes ☐ No

If yes, modality of the synthetic data:
☒ Text ☒ Image ☒ Video ☐ Audio ☐ Other If so, please specify…

If yes, specify the general-purpose AI model(s) used to generate the synthetic data if available on the market:
A range of general-purpose AI models, including third-party vision-language and image generation models, and Luma’s own models.

Information about other AI models, including provider’s own AI model(s) not available on the market, used to generate synthetic data to train the model to which this Summary applies:
Luma also used internal models not available on the market, including teacher models used to distil the models covered by this Summary.

Additional comments (optional):
Synthetic data is used to complement, not replace, real data where content types are under-represented, for example particular visual styles, rare scene compositions and precisely controlled camera movements.

2.6 Other sources of data

Have data sources other than those described in Sections 2.1 to 2.5 been used to train the model?
☒ Yes ☐ No

If yes, provide a narrative description of these data sources and the data:
Content generated through scripted (non-AI) rendering and simulations, and human curation and annotation of training data.

Additional comments (optional):
—

3. Data processing aspects

3.1. Respect of reservation of rights from text and data mining exception or limitation

Are you a Signatory to the Code of Practice for general-purpose AI models that includes commitments to respect reservations of rights from the TDM exception or limitation?
☐ Yes ☒ No

Describe the measures implemented before model training to respect reservations of rights from the TDM exception or limitation before and during data collection, including the opt-out protocols and solutions honoured by the provider or, as applicable, by third parties from which datasets have been obtained:
Luma licenses part of its training data from rightsholders and their representatives under commercial licence agreements. Luma collected online content through targeted crawling of a list of specific websites selected by Luma, rather than a general crawl of the web. Each website was reviewed by a member of Luma’s staff before it was added to the list, and websites known for hosting pirated or otherwise copyright-infringing content were not added. Luma’s crawler did not circumvent CAPTCHAs, paywalls or other technical access restrictions and did not access content that requires a login.

Additional comments (optional):
Luma adopted a written copyright policy under Article 53(1)(c) AI Act on 23 October 2026. Rightsholders may submit complaints concerning Luma’s models as described in Section 5 of Luma’s Terms of Service.

3.2 Removal of illegal content

General description of measures taken:
Websites known for hosting pirated or otherwise copyright-infringing content were excluded from Luma’s crawling (see Sections 2.3 and 3.1). Luma also takes measures to detect and remove child sexual abuse material and terrorist content from training data.

3.3. Other information (optional)

Other relevant information about data processing (optional):
—