Video was the line small firms flinched at. Seedance 2.5 is quietly changing the maths

Every owner knows the moment in the marketing meeting where the plan meets the bank balance. The social posts are cheap. The email list costs next to nothing.

The blog is a weekend’s work. Then someone says the word “video”, and the room goes quiet, because everyone at the table knows what a proper thirty-second clip has always cost: a crew for a day, a location, an editor, and an invoice that makes a founder wince. For most firms the honest answer was to skip it, and then to watch better-funded competitors look more established for no reason other than the size of their production budget.

That flinch is worth writing about, because it is starting to fade, and not for the reasons the usual hype cycle gives. The story is not that “AI can make video now” in some vague futurist way. It is that one specific capability crossed a line this year, and that line happens to sit almost exactly where a small firm’s real needs begin.

The number that actually changed is thirty seconds

The tool behind most of this conversation is Seedance 2.5, the latest video model from ByteDance, the company that owns TikTok, which it unveiled in June at its Volcano Engine FORCE event. Set the launch language aside and the useful fact is dull and specific. From a single written prompt or one reference image, it renders a continuous thirty-second shot at native 4K, with the sound generated in the same pass as the picture rather than added afterwards.

Thirty unbroken seconds reads like a trivial figure until you have tried to make anything usable with the earlier generation of these tools. The first wave of AI video fell apart the moment it ran past a few seconds. Faces drifted, hands rearranged themselves, backgrounds melted, and the clip collapsed before it could say anything at all. Holding a coherent scene steady for a full thirty seconds is not a cosmetic upgrade. It is the difference between a novelty you show a colleague and something you are willing to put a brand name against. Thirty seconds is a pre-roll ad. It is a product explainer. It is the establishing shot of a campaign. It is, near enough, the unit of video a small firm actually buys.

What that does to a marketing budget

Picture the video a typical SME wants across a single month and never commissions, because no one piece can justify a shoot on its own. A short spot for a new product. A looping backdrop for the website’s hero section. A fifteen-second cut to test in paid social. A seasonal teaser for the week before a launch. On their own, none of them clears the bar for hiring a crew, so most owners go quietly without and accept that their channels look thinner than the business really is.

Generating those pieces from a prompt and a handful of reference images folds that whole list into an afternoon. The saving is not only cash, though the cash is real enough. It is that the decision itself changes shape. The question stops being “can we afford to make this” and becomes “is this worth twenty minutes to try”, and a great many ideas that never survived the first question sail past the second.

There is a strategic point buried in that shift. When an attempt costs almost nothing, you can put three visual directions for the same campaign in front of the team before committing, rather than staking the budget on one and hoping. The reference handling matters here too: Seedance 2.5 accepts up to fifty inputs in a single run, so you can feed it your product, your brand palette and a location still and get output that keeps a consistent look across the whole clip, which is exactly the thing small firms usually lose when they cobble content together from whatever is to hand. Take a small homeware shop with a new autumn range. Last year that was one hero photo and a caption. Now it is a warm thirty-second sweep across the display for the site, a square cut for Instagram, and two alternative teasers run against each other to see which lands, all built from the same reference shots in a morning. None of that was affordable a year ago.

The music video is the clearest proof

If you want to see how far the floor has dropped, look at the one form of video that was always the most expensive per second anyone made: the music video. For most of its history it was a rich artist’s privilege, funded by a label that expected to recoup the cost, and independent musicians simply went without. Those musicians are, in every practical sense, small businesses. They run their own release schedule, their own marketing and their own cash flow, with no backer standing behind them.

The distance the technology has covered shows up cleanly in the version numbers, which is the kind of before-and-after a sceptic can actually check. Seedance 2.0, barely a year old, produced clips of four to fifteen seconds at up to 1080p and took twelve reference inputs. Seedance 2.5 lifts that to a single thirty-second shot at native 4K and up to fifty references, and those references can include audio. For a musician that last detail matters more than the resolution, because it means feeding in the actual track and asking for visuals that move with it, instead of describing a mood and hoping. ByteDance also reports around twenty per cent better prompt adherence than the older model, though that figure is the company’s own rather than an independent test.

The upshot is not that an unsigned artist can now summon a Hype Williams video from a laptop. It is that the visual which used to sit permanently out of reach, a proper thirty-second piece with a consistent look and a deliberate camera move, is now something they can attempt on the same afternoon they finish the mix. The logic that frees a musician frees a plumber, a bakery or a two-person software firm just as neatly.

Where it still falls over

None of this replaces a real shoot when the work genuinely calls for one. If an idea leans on a specific human performance, a presenter’s face doing something exact, a choreographed sequence, you still need people in a room and someone directing them. Emotional acting, the flicker of a real expression, is the hardest thing for any of these systems to fake, and it is precisely what a lot of the best video lives on. Anyone selling a prompt as a full stand-in for that is overreaching. The models have tells if you look closely, and hands are still where they most often give themselves away.

There is a cost discipline the demos never mention, either. Length and resolution are what burn through credits, so a full thirty-second 4K render is not free, and going for maximum quality on the first attempt is exactly how you watch a good idea come out wrong at full price. Signing in and the starter credits cost nothing, which is enough to render a first clip and learn how Seedance 2.5 behaves before any real money is involved, but the tap is not unlimited. The workflow that keeps it cheap is unglamorous: draft the shot short and at low resolution, correct one thing at a time, and only pay for the finished 4K version once the cheap draft already works.

The sensible way to file all this is not as the end of video production but as the removal of one particular barrier. The barrier was cost, and it kept the single most persuasive format in marketing out of reach of exactly the firms that most needed to persuade. It is coming down. What a business does with the room that opens up is, as ever, the part no software will do for you.