Updated November 2026.
Streaming services have run on machine learning for more than a decade. Recommendations, artwork, and encoding all use it. Between 2024 and 2026, generative models moved into features viewers can see: search you can talk to, recaps cut from a show's own footage, dubbing for titles that never had one, and ads built around the worlds of the shows.
Netflix put the change in plain words in its April 2026 letter to shareholders: "We've been using machine learning and AI for many years," and GenAI brings "new opportunities." Three months later it reported that GenAI workflows had been used in roughly 300 of its titles.
Below are the main uses on the largest platforms, with dates and sources. Each section ends with what the feature means for the distributors and sales agents who license titles to those platforms.
1. Search and discovery you can talk to
In May 2025 Netflix opened an opt-in beta on iOS that let members search in plain phrases such as "I want something funny and upbeat." By July 2026 it was adding voice search and "AI-powered natural language search". The same letter says Netflix uses large language models "to improve title discovery."
On 3 November 2026 Netflix launched AI Discover, a new tab on its TV app. Members pick a theme such as a mood, a group, or the time they have. They can narrow it, ask for a fresh set, or tell Netflix a title is not for them. Netflix says "AI helps create the themes you see" and that it will explain the results. The tab is rolling out first in the US, to members who use Netflix in English on supported TVs.
Prime Video got there earlier on recommendations. In November 2024 Amazon said its recommendations use generative AI on Amazon Bedrock to find titles with similar plot points and character arcs, beyond shared genres. Personalized artwork is older still. Netflix has long shown different thumbnails to different viewers, such as a Robin Williams image of Good Will Hunting for a comedy fan.
For distributors, a model that matches "funny and upbeat" has to read something. It reads synopses, genres, and tags. A title with a one line synopsis gives it little to work with. Molten Cloud Content Management includes Metadata Enrichment, which suggests missing synopses, genres, and identifiers from external sources for your team to approve.
2. Recaps cut from the show itself
Prime Video launched text based X-Ray Recaps in 2024. On 19 November 2025 it went further with Video Recaps. Generative AI analyzes a season's key plot points and character arcs and picks the clips that matter. It pairs them with audio effects, dialogue snippets, and music, and an AI generated narration ties the recap together.
The beta covers select English language Prime Original series in the US, on living room devices first. A recap button appears on the detail page when a viewer opens the next season.
For distributors, the question comes later. Today the feature runs on series Amazon owns. If platforms extend machine made recaps to licensed series, each recap is a new edit of licensed footage. Check whether your licences let the platform cut promotional or derivative material, and who approves it.
3. Dubbing for titles that never had one
In March 2025 Prime Video began an AI aided dubbing pilot in English and Latin American Spanish on 12 licensed movies and series. Amazon limited it to titles "that would not have been dubbed otherwise." It called the pilot "a hybrid approach to dubbing in which localisation professionals collaborate with AI to ensure quality control."
Netflix applies AI to the text track. Its January 2026 letter says it is "using AI to improve subtitle localization, making it easier for our titles to reach more viewers around the world." The audience is there. Non-English viewing is now more than a third of all viewing on Netflix, up from under 10% a decade ago.
Dubbing is where AI meets licence terms most directly. A machine dub creates a language version that may not have existed when the deal was signed. Read the language clauses before a platform proposes one. You need to know which languages each licence grants and who owns a new audio track.
4. Ads built from the shows
At its May 2025 upfront, Netflix said generative AI would "instantly marry advertisers' ads with the worlds of our shows." The first formats are interactive midroll and pause ads, due in all of its ad-supported countries by 2026. The ad plan then had more than 94 million monthly active users worldwide.
The January 2026 letter says Netflix began testing AI tools "to help advertisers create custom ads based on Netflix's intellectual property." By July it had expanded AI-powered tools "across the full advertising lifecycle." Revenue is growing more slowly than Netflix planned. In October 2026 it lowered its 2026 ad revenue outlook to about $2.7 billion from about $3 billion, citing a slower rollout of some ad product capabilities. The ad tier comes to 15 more countries in 2027.
For distributors, every new ad tier adds revenue lines tied to impressions and ad share. Those lines arrive in platform reports, each in its own format. See how ad based royalties are tracked at scale and how the 2025 numbers moved toward AVOD and FAST.
5. Picture and sound quality
On 29 October 2025 YouTube introduced Super Resolution on its TV apps. It uses AI to upscale videos uploaded below 1080p to HD. YouTube keeps the original files and aims to support 4K upscaling later. Creators can turn it off.
Prime Video's Dialogue Boost uses AI to analyze a title's original audio, find where speech is hard to hear, and lift it. Amazon said in 2024 it works in seven languages. Under all of this sits adaptive bitrate streaming, which adjusts quality to each viewer's connection so playback keeps going when bandwidth drops.
For rights holders, an upscaled stream is a version of the picture you did not deliver. YouTube gives creators an opt out. Ask your other platforms whether they offer the same.
6. Generated footage inside the shows
In July 2025 Netflix co-CEO Ted Sarandos said El Eternauta used generative AI for a building collapse in Buenos Aires. Sarandos called it "the very first Gen AI final footage to appear on screen in a Netflix original series or film." The sequence was finished 10 times faster than traditional methods allowed.
A year later Netflix reported GenAI workflows in roughly 300 titles: "In some cases, productions would have had to leave out key shots and sequences in the absence of GenAI technology." In March 2026 Netflix acquired InterPositive, the filmmaking tools company founded by Ben Affleck. We cover the production side in AI in VFX.
For anyone buying or selling finished films, this adds a diligence question: which shots are generated, with which tool, and trained on what. Ask it of your own productions before a buyer does.
7. AI streaming services
Some services are built on generated content from day one. Fable's Showrunner went wide on 30 July 2025, the same day Amazon made an undisclosed investment in Fable. Users type prompts and the platform generates scenes and full episodes, with voices, dialogue, and camera shots. CEO Edward Saatchi calls it the "Netflix of AI." Saatchi also said its terms make clear that "the model provider owns everything that's generated with the model."
The largest studio deal in this space did not last. On 11 December 2025 Disney and OpenAI announced a three year licence for more than 200 Disney, Marvel, Pixar, and Star Wars characters on Sora. Curated fan videos were planned for Disney+, with a $1 billion Disney equity investment in OpenAI. The agreement excluded "any talent likenesses or voices." On 24 March 2026 OpenAI said it would discontinue Sora, and Disney exited the deal.
Rights holders can take two points from these cases. Ownership of generated output is set in the platform's terms. A character licence can leave out the actor's face and voice. Both belong in the contract. For a format that grew up on AI, see our analysis of microdrama.
What to check in your contracts
| AI feature | Live example | Contract question |
|---|---|---|
| AI dubbing | Prime Video pilot on 12 licensed titles (2025) | Which languages does the licence grant, and who owns a new track? |
| Machine recaps and clips | Prime Video Video Recaps, beta (2025) | May the platform cut promotional or derivative edits, and who approves them? |
| AI ad formats | Netflix interactive midroll and pause ads (2026) | Can the title's world, characters, or footage appear in advertiser creative? |
| Upscaling and enhancement | YouTube Super Resolution (2025) | May the platform alter the delivered picture or sound, and can you opt out? |
| Generated footage | El Eternauta (2025), about 300 Netflix titles (2026) | Which shots are generated, with what tool, and who cleared the training data? |
| Conversational discovery | Netflix AI Discover (2026) | Does your metadata give the model enough to match? |
Where Molten Cloud fits
Molten Cloud is a rights management and royalties platform for film and TV distributors. Rights Management holds what is licensed, to whom, for which territories and windows, and on what terms. Custom fields are indexed, so a team can add a field such as "AI dubbing permitted" and filter the whole catalog by it. AI Rights Management reads contract PDFs and drafts the rights data for someone on your team to review.
Royalties Management automates royalty and participation calculations and reporting, including FAST and AVOD revenue. Content Management covers storage, delivery, and screening lists.
To see your catalog's terms in one searchable place, book a demo of Molten Cloud.




