# Luma added Modify Video to Dream Machine with three transformation strengths Luma added Modify Video to Dream Machine on June 4, 2025, letting users transform footage while choosing Adhere, Flex or Reimagine behavior for source-video fidelity. Canonical: https://rangoon.ai/insights/archive-luma-modify-video-june-2025/ Author: Rangoon Editorial (https://rangoon.ai/insights/editorial/) Published: 2026-10-24 Updated: 2026-10-24 Event date: 2025-06-04 Tags: [Luma](https://rangoon.ai/insights/tags/luma/), [video AI](https://rangoon.ai/insights/tags/video-ai/) Source note: Historical reporting: Source 0 is Luma's dated Modify Video launch record. Source 1 is Luma's technical guide, used for the later API parameters and mode definitions rather than date proof. Image: Modify Video fidelity choices from clip to edit. Original Rangoon explanatory diagram: Source boundary: dated primary release reporting; original editorial analysis. No vendor interface, photograph, measurement result, or endorsement implied. ## Key takeaways - Luma dated the Dream Machine Modify Video launch to 2025-06-04. - The release exposed Adhere, Flex and Reimagine choices, which set different relationships between the source clip and its transformation. - The dated launch used Ray2, accepted footage up to 10 seconds and listed Web and iOS delivery. ## A video edit started from recorded performance [source 1](https://lumalabs.ai/changelog/introducing-modify-video) Luma's June 4, 2025 changelog introduced Modify Video in Dream Machine as a way to transform an existing clip without throwing away its motion and performance. The event page describes extracting dance movement, facial expression and choreography from footage, then mapping those signals to different characters or scenes. It also lists smaller changes: a wardrobe, prop or background could be altered while the rest of the shot stayed in place. The important change was therefore not just a new image style. It offered a video-to-video operation in which the source clip remained an input that constrained the result. Luma framed the surface around reimagining environments, lighting, texture and scenario while preserving parts of the original performance. Those are Luma's product claims. They do not show that every edited clip preserves a performer accurately, nor do they answer consent, likeness or clearance questions for the supplied footage. ## Adhere, Flex and Reimagine made the tradeoff explicit [source 1](https://lumalabs.ai/changelog/introducing-modify-video) [source 2](https://docs.lumalabs.ai/docs/modify-video) The launch was unusually specific about the kind of control it exposed. Adhere was for modest lighting and texture work close to the original. Flex allowed a larger style move while retaining recognizable elements. Reimagine was the setting for broad scene changes. The current Luma technical guide retains the same family of modes as adhere_1 through adhere_3, flex_1 through flex_3 and reimagine_1 through reimagine_3. In the guide, Adhere stays close to the source, Flex permits more change while keeping shapes, characters and details recognizable, and Reimagine can loosen that relationship enough to change the world, style or even characters. This is a more useful description of the feature than an undifferentiated claim of video editing: the mode selection is where a user declares how much of the input should survive the transformation. ## The June surface had concrete duration, model and format bounds [source 1](https://lumalabs.ai/changelog/introducing-modify-video) [source 2](https://docs.lumalabs.ai/docs/modify-video) At launch, Luma told users to upload footage of up to 10 seconds, choose a Modify Strength Level, describe the intended result and generate. Its technical-details block named Ray2 as the model, Web and iOS as the platforms, 720p and 1080p output, and .mp4, .mov and .wmv input formats. It also said modified videos retain their original audio, with an option to add audio through a separate feature. The present technical guide describes an API and SDK version with a prompt, a required video URL, an optional preferred first-frame URL and named model identifiers including ray-2 and ray-flash-2. That later guide helps explain the input contract, but it does not rewrite June 2025 availability or pricing. The launch page is the historical authority for what Luma said it released that day. ## A transform control is not proof of faithful editing [source 1](https://lumalabs.ai/changelog/introducing-modify-video) [source 2](https://docs.lumalabs.ai/docs/modify-video) Modify Video shifted the creative operation from replacing a shot wholesale to choosing how tightly a transformation should follow one. That can be valuable for previs, tests and alternative treatments because an existing camera path and performance provide structure. It does not make the generated result a verified version of the input clip. Luma's documentation describes how each mode is intended to behave, but it gives no universal guarantee of identity persistence, temporal stability, legal permission or semantic accuracy. The historical event supports a June 4 launch, a Dream Machine workflow, named strength levels and Ray2-era constraints. It should not be read as evidence that motion capture equipment was no longer needed for every production, that a transformation is safe for publication, or that uploaded material can be reused without rights review. Those judgments depend on the footage, setting, terms and human review outside the dated product announcement. ## Sources 1. [Luma: Modify Video launch](https://lumalabs.ai/changelog/introducing-modify-video) 2. [Luma: Modify Video technical guide](https://docs.lumalabs.ai/docs/modify-video) ## Related reading - [Luma released Ray2 for paid Dream Machine subscribers](https://rangoon.ai/insights/archive-luma-ray2-january-2025/) - [Luma launched Ray3 with Draft Mode and ACES EXR HDR output](https://rangoon.ai/insights/archive-luma-ray3-september-2025/) - [Luma added Camera Motion Concepts to Ray2](https://rangoon.ai/insights/archive-luma-camera-motion-concepts-march-2025/) ## More information - [Documentation index](https://rangoon.ai/llms.txt) - [AI access and policies](https://rangoon.ai/ai/) - [Source repository](https://github.com/hypler-dev/rangoon)