Flora adds Seedance 2.5, ByteDance's 30-second video model
By Steven Van ·
The integration takes up to 50 references at once, spanning images, video and audio, in a single generation.
Seedance 2.5 has arrived on FLORA, the node-based infinite canvas for creative teams, and the integration leans into everything that makes ByteDance's new flagship different: 30-second generations with multi-shot storytelling, reference input of up to 30 images, 10 videos, and 10 audio clips in a single pass, sharper motion, and better character consistency. For a platform built around chaining models together into workflows, a video model that takes 50 references at once is close to a native fit.
Multi-shot storytelling in one generation
The capability that separates Seedance 2.5 from the previous generation is not just clip length, it is structure. A 30-second generation with multi-shot storytelling means the model can cut between angles and beats inside a single output, rather than producing one continuous take you then have to assemble with other clips. That collapses a chunk of the edit into the generation itself: establish, cut in, react, resolve, all from one prompt and one reference set. On FLORA's canvas, a single node now does what previously took a row of generations plus manual sequencing.
30 images, 10 videos, 10 audio clips in one pass
The reference budget is the other headline. FLORA's integration accepts up to 30 images, 10 videos, and 10 audio clips in a single pass, 50 inputs total. In practice, that is enough to pin down a character from multiple angles, a wardrobe, a product, a location, a motion reference, and a soundtrack simultaneously. Heavy referencing is how you get from "AI video that looks cool" to "AI video that matches the brief," and it is precisely the kind of input assembly FLORA's node graph is good at: your image generations, uploaded footage, and audio can all feed the Seedance node directly from elsewhere on the canvas.
Sharper motion and character consistency, the unglamorous wins
FLORA also calls out sharper motion and better character consistency, and these are the improvements that decide whether a model survives professional use. Character drift across shots has been the single biggest tell of AI video and the single biggest cost in retries. ByteDance's official material for Seedance 2.5 emphasizes smoother, more consistent motion and deeper understanding of reference footage, capturing framing and cinematic intent rather than just copying movement. Combined with the 30-image reference ceiling, consistency stops being luck and becomes something you can engineer.
Why a canvas platform benefits most from a reference-hungry model
FLORA aggregates more than 50 text, image, video, and audio models into one connected workspace, and its whole thesis is that creative control comes from wiring outputs into inputs. A model that accepts 50 multimodal references rewards exactly that workflow. Where a chat-style tool makes assembling 50 inputs a chore, on FLORA the reference set is just the upstream half of your graph. Seedance 2.5 is rolling out across many platforms this week, but the reference-stacking workflow is where a canvas-first tool gets to show why it exists.
Seedance 2.5 is available on FLORA now. If you already keep character sheets, style frames, and audio on your canvas, point them at the new node and see how far one pass gets you.
Sources: FLORA announcement on X, FLORA official site, ByteDance Seed: Seedance 2.5