Synthesia on X

Syren is the draw. People describe starting from a prompt, a document, a link or an existing deck and getting motion graphics, avatars, voiceover, music and B-roll in the brand's own style, learned from videos the brand has already made. Chat-based refinement gets singled out, as does pacing control, which one person calls "the real unlock". The claimed payoff is time: brand video going from weeks to minutes, launch videos no longer a two-week project, one prompt turning into a shipped asset.

Live Sessions avatars are the second thread, level with the head-to-head avatar poll at 9 and 9 people. People praise Sessions as practice for sales roleplays and difficult conversations — a journalist scored 57% on a promotion talk with an AI employee — and an interactive avatar of a cartoon Jensen Huang is described as a glimpse of what Sessions can become. In the poll, several say they would not have spotted the avatar without the voice or another tell.

Smaller groups are simply enthusiastic (7 people) or eager to try it, drawn in by free credits (4 people).

Most of the complaints are about small physical tells. Watching a side-by-side test, people name lip-sync that is slightly off, a voice that sounds monotone, no pause to take a breath between sentences, hands moving more than a person's would, and a nagging nod of the head; one spots an inverted blink easing and another says they knew in a fraction of a second. The same posts often concede the output is good otherwise. One viewer says they tune out any video with no human in it.

Cost is the next thing named, by 2 people: one asks the team about "USD 150 for only two videos", and another tracks Syren's promo credits being cut on paid plans within a fortnight of launch. As many again hit missing pieces — no .md script files accepted as input, and an avatar render that came out pixelated — and 2 are outright skeptical, calling AI video "another slop machine" and warning that AI-written scripts make avatar videos obvious to viewers.

The biggest group is weighing it against doing the work by hand or hiring for it. One person who already makes videos with AI, and spent a while getting the file, brand and layout setup right, sees in Syren "an easier approach to doing this for the knowledge worker" and wants to try it. Two more call it a solid option for teams putting out tutorials and feature updates every week; another says it could be useful for small teams with no video editor on hand.

The other two comparisons are one voice each. One wonders whether it can match the style of videos cut in traditional editing software. The other puts it in a field with HeyGen, Runway, Kling and CapCut and says the best tool depends on the job — judged on control, consistency across shots and how the output fits the rest of the editing workflow, "not only on a single impressive generation".

Most of the questions are conditions rather than objections. People want to know how it handles brand specifics, how much control they get over the final result, and whether the quality of the output will hold up — one puts it plainly: the workflow sounds good, "but the quality of the final output will decide everything". One wants to see what happens when it is handed a messy document.

The consistency question comes up in its own right: whether the style holds across something like twenty videos or only nails the first, and the view that making one good video is easy while keeping the next twenty consistent is the hard part. Someone replied to that one that it does hold, which is their claim, not a demonstration.

Separately, a reader of the Sessions sales-roleplay thread asks whether the AI coaching still holds up when the rep is actually decent, "or does it mostly catch the obvious stuff".

The wish is about making the trial answer the buying question, not just show the product. The person suggests the free trial begin with a client brief already loaded in Claude, so an agency can generate a draft it could genuinely put forward for approval — which turns the trial into a test of whether they would trust it on a paid project. It was raised in the thread about Sessions and sales roleplays, addressed to a co-founder, and frames the decision as concrete rather than exploratory: would this hold up on work a client is paying for. Nobody else in the posts asks for the same thing.