Shotstack adds Luma's Ray 3.2 and Uni-1 to its render API
The four Luma models run on a Shotstack API key and credits inside the same render that adds titles, captions and music.

Key takeaways
- Shotstack added four Luma models on November 19, 2026: Ray 3.2 Video to Video, Ray 3.2 Reframe, Uni-1 and Uni-1 Edit.
- Ray 3.2 makes 5- or 10-second clips and outputs up to 1080p, but vertical reframes at 9:16 and 3:4 top out at 720p for now.
- All four models run on the customer's Shotstack API key and credits, in the same render that adds titles, captions and music.
Shotstack now runs four of Luma's models inside its video render API, so a developer can restyle footage, reframe it for another channel and generate stills that match, all inside the same render that adds titles, captions and music. Shotstack announced the addition on November 19, 2026.
The four models are Luma Ray 3.2 Video to Video, Luma Ray 3.2 Reframe, Luma Uni-1 and Luma Uni-1 Edit. Three of the four start from something a creator already has, a clip or an image, instead of a blank prompt. That is the part worth understanding: the other video models in Shotstack's catalog make new footage from a prompt or a start image, while these three change footage a shoot already produced.
What Shotstack added
Shotstack sells a video editing API. Developers send it a JSON timeline of clips, titles, captions and music, and it renders a finished video. Luma's models now sit inside that pipeline as assets, called by their own model IDs:
- Luma Ray 3.2 Video to Video (
ray-3.2-video-to-video) changes how a clip looks and keeps its motion and performance as shot. - Luma Ray 3.2 Reframe (
ray-3.2-reframe) changes a clip's shape, such as 16:9 to 9:16, and fills the new edges from your prompt. - Luma Uni-1 (
uni-1) creates a 2K image from a prompt, in nine aspect ratios from 3:1 to 1:3. - Luma Uni-1 Edit (
uni-1-edit) edits an image you already have, guided by up to eight references.
All four run on the customer's Shotstack API key and Shotstack credits. A developer can read each model's options from GET /edit/v1/models?expand=options.
How Ray 3.2 Video to Video restyles a clip
Ray 3.2 Video to Video takes a clip and a prompt and changes what the viewer sees, not what happens. The camera move, the timing and the performance stay. The setting, the light and the style are whatever the prompt describes.
Shotstack's walkthrough uses a 5-second AI-generated clip of a concept car on a coast road in morning sun and restyles it twice, once in fresh snow and once at night, with the camera move and the car's position unchanged. Each version is a variant, a prompt plus a start image, and Shotstack's sample code builds one edit per variant and sends each to the render endpoint.
Two options control how far the result drifts from the source. startSrc is a start image used for the first frame, and strength sets how much room the model has. The adhere levels hold motion, composition and structure closely, while flex and reimagine give the prompt more room.
How Ray 3.2 Reframe changes a clip's shape
Reframe changes a clip's aspect ratio without cropping. In Luma's words, the source video content is preserved frame-for-frame, and the model generates new footage to fill the canvas around it.
Shotstack shows the difference with numbers. Cropping a 1920 by 1080 clip to 9:16 keeps a strip 608 pixels wide, less than a third of the picture, and throws the rest away. Reframe keeps the whole frame and fills the new edges from a prompt.
In an edit, a reframe is one clip with a target aspectRatio. Give the output the same ratio and the finished video comes out vertical.
What Uni-1 and Uni-1 Edit do
Uni-1 makes 2K stills from a prompt in nine aspect ratios. Shotstack suggests them for the frames around footage: an opening frame, a background for a price card, an end card.
Uni-1 Edit starts from an image the creator already has, such as a product photo or a packshot. That image goes first in imageUrls, up to eight references follow it, and the prompt describes the change. Luma says the edit preserves composition and structure unless told otherwise, so a product stays recognizable in a new scene. Shotstack tested it by placing one frame from the car clip into four scenes, a studio, a salt flat, a car park and a city street, and the car's shape, paint and light bar carried over each time.
What you need and what it costs
You need a Shotstack API key. The four Luma models run on that key and Shotstack credits, and Shotstack bills generations from the credits at its published AI generation pricing.
Generation does not have to happen inside a render. A developer can send an asset to POST /edit/v1/generate, poll GET /edit/v1/generate/{id} until the status reads done, then use the returned URL as a normal src in as many edits as needed. Shotstack says repeated generations are cached: an edit that includes an asset already generated with the same model, prompt and options reuses the file at no charge, so re-rendering after a headline change costs the render and not a second generation.
The limits to know
Ray 3.2 makes clips of 5 or 10 seconds. Reframe takes a source of up to 10 seconds. Shotstack generates 5 seconds when the source is 5 seconds or shorter and 10 seconds when it is longer, and bills for the generated length. Longer videos are meant to be processed shot by shot.
Output is 360p, 540p, 720p or 1080p, with 720p as the default. Vertical reframes at 9:16 and 3:4 top out at 720p for now, so a creator who needs a full-HD vertical cut from a reframe cannot get one yet.
What it means for creators
The set is aimed at volume, not at a single hero shot. Luma lists the jobs Ray 3.2 Video to Video is built for: a new setting, relighting, a product swap, and adapting one ad for several markets. Each one used to mean another shoot or hours in an edit suite.
"Every platform wants a different shape," Derk Zomer, co-founder and CEO of Shotstack, said in the announcement. "With Ray 3.2, our users stop cropping their best shots to fit." Chris Swadling, AI GTM Lead at Luma AI, said the partnership brings Luma's image and video models "directly into automated content production workflows."
A creator can also restyle a clip generated with another model. Shotstack says a video made first with Seedance, Wan or any other video model on its platform can be used as the src of a Ray 3.2 clip. What the pages do not say is whether the two-stage generation costs less than generating the final look in one pass, or how well Reframe holds up on footage with fast motion. Neither has been published.
Shotstack offers a free API key for its sandbox. The four Luma models run on the same key and credits as the rest of the Edit API, billed per Shotstack's AI generation pricing.
Sources
- shotstack.io - primary; the four models, model IDs, clip lengths, resolutions, the Luma and Shotstack quotes