Answers · AI model licenses
Open weights vs open license: can I use an open-weight model's outputs commercially?
Last updated
TL;DR: Downloadable weights tell you nothing about what you may do commercially; the license file does. Many open-weight models are under custom community licenses, not open source licenses, and they add conditions that apply even when you only use the outputs. Two common patterns: revenue thresholds and user thresholds. Lightricks' LTX-2.x Community License (August 11, 2026) is free for commercial and production use, but entities with annual revenue of at least US $10,000,000, counted across affiliates, need a paid Commercial Use Agreement for anything beyond non-commercial testing and research. Meta's Llama 4 Community License allows commercial use, but companies whose products had more than 700 million monthly active users must ask Meta for a license, and anyone who uses Llama outputs to train a model they distribute must start its name with "Llama". LTX-2.x says Lightricks claims no rights in your outputs, yet no use of them may break the agreement; Llama 4 attaches its naming duty to outputs you train on. A permissive license such as Apache-2.0 has no such conditions. The Open Source Initiative's definition counts a model as open source only if its weights, code and data information come under terms that let anyone use it for any purpose without asking permission.
What should I check, step by step?
1. Find the license for the exact checkpoint: the license field and LICENSE file on the model card, not the family name or a blog post. Lightricks, for example, has a LICENSE-2 (January 5, 2026) for LTX-2 releases up to LTX-2.3 and a LICENSE-2_x (August 11, 2026) for LTX-2.5 releases.
- Check for a revenue threshold and how it is counted: LTX-2.x counts subsidiaries, affiliates and companies under common control together, and a Commercial Entity's revenue-generating use, use that affects end users, or training for commercial use is not non-commercial.
- Check for a user threshold: Llama 4 requires a separate license from Meta if your products had more than 700 million monthly active users in the month before the release date.
- Check what the license says about outputs: Llama 4 makes you name a distributed model trained on Llama outputs "Llama ..."; LTX-2.x bans using Outputs, for commercial use, to train other models except permitted LTX-2.x Derivatives.
- Check territorial limits: the Llama 4 Acceptable Use Policy does not grant the license rights for its multimodal models to individuals domiciled in, or companies with a principal place of business in, the EU (end users of a product built on them are not affected).
- Check duties when you ship: Llama 4 requires "Built with Llama" displayed prominently and a copy of the agreement; LTX-2.x requires passing on its use restrictions, keeping watermarking and AI-disclosure features intact and complying with AI transparency laws.
- Fine-tunes and LoRAs inherit the base license: LTX-2.x says a Commercial Entity must get a paid license before using any Derivative, whoever made it.
- Record the model, version, license file and your revenue or user tier on the date you generated the output, so you can show which terms you relied on.
Why "open weights" is not "open source"
The Open Source AI Definition 1.0 requires that a system be usable for any purpose without having to ask for permission, and that weights, code and training data information be available under OSI-approved terms. A license that requires a paid agreement above a revenue line or a user count, or that bars some uses of the outputs, does not meet that test, even though anyone can download the weights.
- That is why the license field matters more than the label: Apache-2.0 and MIT models carry no revenue, user-count or output-naming conditions, while community licenses can.
LTX-2.x: the US $10M revenue line
Section 2.1 of the LTX-2.x Community License grants use for any purpose subject to Attachment A, but Entities with annual revenue of at least US $10,000,000 must obtain a paid Commercial Use Agreement for any use other than a Non-Commercial Purpose. Unlicensed use by such an entity is a material breach, and the licensee owes the license fees for the period used at Lightricks' standard rates.
- Section 5 says Lightricks claims no rights in the Output you generate, but no use of the Output can contravene the agreement. LTX's licensing FAQ says the US $10M threshold is total annual company revenue measured across your organization, not revenue from the product that uses the model.
- The license also terminates automatically on unauthorized commercial use by a Commercial Entity, and Lightricks may revoke it if a model is modified so it can no longer include required AI disclosures.
Llama 4: the 700 million user line
The Llama 4 Community License (effective April 5, 2025) grants a royalty-free license to use, reproduce, distribute and modify the Llama Materials. Section 2 adds that if products made available by or for you or your affiliates had more than 700 million monthly active users in the preceding calendar month on the release date, you must request a license that Meta may grant in its sole discretion.
- Section 1.b.i requires "Built with Llama" on a related website, interface or documentation when you distribute or make available a product that contains Llama Materials, and a "Llama" prefix on any distributed model created or improved with Llama outputs. Use must follow the Llama 4 Acceptable Use Policy.
What do the license checks say?
Yes means the license grants the use, on the duties listed. Duties such as source_disclosure, share_alike and network_source_offer are the ones that conflict with keeping a product closed-source.
| Use | Answer | Why |
|---|---|---|
| An Apache-2.0 open-weight model run yourself for commercial output (Apache-2.0) | Yes | Every use you described is granted by Apache 2.0, and nothing you described is prohibited. Duties: attribution, license_notice, state_changes. |
| A Llama 3.3 model (Llama 3.3 Community License) run yourself for commercial output (LicenseRef-Llama-3.3-Community) | Yes | Every use you described is granted by Llama 3.3 Community, and nothing you described is prohibited. Duties: attribution, license_notice, acceptable_use_policy, model_name_prefix, user_threshold. |
How do I ask Rightshold about this (free, no key)?
Add the MCP server https://www.rightshold.com/mcp to Claude, Cursor, ChatGPT, Grok or VS Code, then ask: “We're a startup self-hosting an open-weight video model for a paid app. Which license does this checkpoint use, and is commercial use allowed?”
{
"tool": "check_license",
"arguments": {
"source_url": "<the model's Hugging Face page>",
"intended_use": {
"commercial_use": true,
"deployment": true
}
}
}- ask_rightshold takes the same question in plain English and tells you which license and use it understood.
- issue_rights_receipt (with an API key) binds an output's SHA-256 to the model, license basis and intended use you record, signs it and commits it to the ledger, so you can show later which terms applied when you generated it.
- check_rights_batch checks every model, image, font and package in a project in one call before you ship.
Rightshold reads the model licenses you point it at. It does not know your revenue, user counts or any paid agreement you hold with a model's licensor. Not legal advice.
What are the sources?
- Lightricks: LTX-2.x Community License Agreement (August 11, 2026)
- Lightricks: LTX license overview (which license applies to which release)
- LTX Licensing FAQ (US $10M threshold is total company revenue)
- Meta: Llama 4 Community License Agreement (effective April 5, 2025)
- Meta: Llama 4 Acceptable Use Policy (EU multimodal restriction)
- Open Source Initiative: The Open Source AI Definition 1.0
Related
- Can I use a Hugging Face model commercially?
- Can I use FLUX or Stable Diffusion images commercially?
- Can I use ChatGPT output commercially?
- Who owns code written by an AI coding agent?
Not legal advice. JSON · Markdown · All answers
Get a free API key in one call
No card and no email needed. 100 free credits; the license and policy checks stay free without a key.
curl -s -X POST https://www.rightshold.com/v1/accounts/signup -H "Content-Type: application/json" -d '{"accept_terms":true}'
export RIGHTSHOLD_API_KEY=rs_... # the api_key from the reply (shown once)
curl -s https://www.rightshold.com/v1/licenses/check -H "Authorization: Bearer $RIGHTSHOLD_API_KEY" \
-H "Content-Type: application/json" -d '{"license":"MIT","intended_use":{"commercial_use":true}}'