AI video news, November 7: agents write the story

Four releases landed inside a day: Higgsfield turned ad spots into short dramas inside Ads Studio, Taiwan's imCore showed Core Studio with 24 agents covering the path from script to colour grade, Wizstar shipped its first reference-to-video model, and ElevenLabs put a native speaker inside the dubbing check. Each item carries a line on why it matters, and the verdict is the same for all four: an agent makes the draft, people make the calls.

A ring of nine small agent figures feeds one vertical frame while a human hand stamps approval
01

Micro Drama Ads: a story-shaped ad inside Ads Studio

On November 6 Higgsfield added a Micro Drama Ads mode to Ads Studio. The blog describes the shape plainly: a character, a problem, and the product as the turn in the plot. An agent in the chat builds the spot from a product description or a finished script. Inside sit 8 styles and 44 ready-made story cards, 18 of them in the Micro drama style, all vertical 9:16.

Billing runs on seconds: 3 credits a second at 480p, 7 at 720p, 16 at 1080p. On the 1000 credit pack a minute at 1080p comes to about $47. The same job runs from Claude or ChatGPT through the /microdrama-ads command over MCP.

Why this matters: ads picked up ready-made drama inside an interface that used to hold a clip generator. A story with three turns now comes out of one conversation, and the entry point drops to a product description.

02

Core Studio: 24 agents standing in for a crew

On November 7 at Taiwan Creative Content Fest in Taipei, imCore Corp. opened Core Studio. It is an AI film production platform that carries a project from script to colour grade, with more than 24 specialised agents splitting the work. The company calls them a virtual crew.

Story Atlas breaks a script into scenes, shots and a storyboard. Then Real Engine takes over with three blocks: Real Video, Real HDR expanding 8 bit to 16 bit, and Real Revive. Every generated result keeps a log of models, prompts, parameters, cost and the outcome of a human check.

Why this matters: a log per shot turns AI production into an auditable process. Once it is visible which model produced the bad take and what it cost, a conversation with a client about quality moves from taste to numbers.

Pipeline diagram: script, storyboard, three render blocks, a per-shot log and a human stamp at the end
03

Wizshot 1.0: up to 9 references per generation

On November 6 Wizstar announced Wizshot 1.0 through PR Newswire, its first reference-to-video model. It takes up to 9 reference images, renders 720p in landscape and in portrait, runs 5 to 15 seconds long, and returns up to 4 videos per generation. Pricing is 5 credits a second.

The company claims it costs 72% less than Seedance 2.0 and that quality is comparable to Wan 3.0 and MiniMax H3. No independent measurements existed as of November 6, so the 72% figure stays a vendor promise.

Why this matters: nine references cover a character, a product and a look in one request. For a series of vertical versions that means less manual swapping between shots, though face consistency still needs a check.

04

Ads Engine: a native speaker checks the dub

On November 6 ElevenLabs described in its blog how the human check works inside Ads Engine, part of ElevenCreative. Dubbing v2 produces the voice track and keeps emotion, timing and timbre, then a native speaker watches the result. Linguists and localisers from ElevenProductions sit in the loop together with creatives from ElevenLabs.

Why this matters: machine dubbing has sounded smooth for a while and still fails on intonation and local idiom. A human in the loop removes that class of error, and a foreign-language cut of an ad gains a clear acceptance gate.

Nine reference thumbnails merge into one vertical generation that fans out into four video clips
05

AI video news: four releases in one day

One line for context: on November 6 Luma published a comparison of clip length in a single pass across 10 models, with Seedance 2.5, Wan 3.0 and Kling 4.0 holding 30 seconds. The details live in a separate piece from this issue.

What the four stories share is one shift: an agent stopped being a button labelled make video. Higgsfield hands it the plot of an ad, imCore hands over the whole pipeline with logs, Wizstar hands over references, ElevenLabs keeps the last word with a person. The draft gets cheaper and acceptance gets more expensive, and those are two different lines in a budget.

06

What to test on your own scenes

One: build a single 15 second micro drama at 1080p and count what a run of five versions costs. Agents love offering options, and the bill grows on the search.

Two: push your own script through Story Atlas and see whether it splits on meaning. Splitting on paragraphs gives even shots and a dull cut.

Three: feed Wizshot 9 references and check whether the character keeps the same jacket and hair between shots. That is the first thing to fall apart in a series.

Four: listen to a dub with the reviewer and without. The difference shows up in the intonation of a question, with the same words on both tracks.

07

AI video news: how this lands in client work

For advertising the micro drama shape is handy. Our vertical video ads ship as a set of versions, and a plot with a turn holds attention for the first seconds, while a viewer decides whether to scroll on.

A short piece with a person in frame is easier on a normal pipeline. Our handcrafted videos from $190 come with light, expression and cut checked before delivery. AI video production running in batches lives on a different sum, up to $2.40 per 1000 views, where the price of a second decides the quote and one story does not.

One more habit: compare a day with the one before it. The November 6 roundup covered Characters on four H100s, Gemini Omni 1.2 Flash and a voice in Muse. The top of the table moves inside a day, so price and a human check are steadier references right now than a place in a ranking.

frequently asked

01What does a minute of Micro Drama Ads cost
About $47 for a minute at 1080p on the 1000 credit pack. Credits run by the second: 3 at 480p, 7 at 720p, 16 at 1080p.
02Do I need a script to make a spot
No, an agent works from a product description and a character. A script gives a tighter result, because the model follows prepared turns and does not invent its own problem.
03Is Core Studio a finished product
It is an imCore Corp. platform with 24 agents covering the path from script to colour grade, launched on November 7 at Taiwan Creative Content Fest. The announcement carried no pricing.
04Is Wizshot 1.0 really 72% cheaper than Seedance 2.0
That is what Wizstar states in its November 6 release. No independent price or quality measurements existed on that date, so the figure is worth testing on your own scenes.
05Who checks the dubbing in Ads Engine
A native speaker from ElevenProductions alongside creatives from ElevenLabs. Dubbing v2 keeps emotion, timing and timbre, and the check happens after generation.