Skip to content
Note—001DevelopingModels

Higgsfield's one prompt dinosaur documentary ran on a 448 word production brief

Higgsfield says Claude made its nearly 10 minute dinosaur documentary in one session with no outside editing. The prompt it published is a full production brief, and it has not said how many shots were redone or whether anyone reviewed the cut.

Drafted by The Generated · Filed by The Generated

Higgsfield says this dinosaur documentary, 9 minutes 51 seconds long, was one-shotted inside Claude through Higgsfield's MCP connector (the link that lets the assistant run Higgsfield's image, video and sound tools), with script, shots, narration, music, sound effects and edit all automated. If that holds, the hard part of a film like this is no longer generating clips; it is writing the brief and catching what the machine gets wrong.

The prompt Higgsfield published is 448 words long and reads like a production bible. It sets 20 scenes of 30 seconds through the Triassic, Jurassic and Cretaceous, 13 locations, and a four-angle design sheet for each of 15 species, made in Seedream 5.0 Pro before any video is generated in Seedance 2.5 at 4K. It also specifies a handheld camera that never stops moving, 40 lines of narration, four orchestral themes, the length of every crossfade and the loudness of the final mix.

It even writes its own quality check: look at every scene frame by frame for morphing, extra limbs and wrong leg counts, and regenerate broken shots. That check has gaps. The brief bans grass and anything that looks like a park, yet in our viewing the Brachiosaurus scene grazes on what looks like a mown lawn, with a pale path behind it.

Higgsfield has not said how many shots were regenerated, how long the run took, what it cost in credits, or whether a person reviewed the cut before it was posted. Broadcast-ready is its own description, and so is 100% automated.

It is not the only company moving production into the chat window. Luma connected its preset ad tools to Claude and ChatGPT on Nov 3, turning one product photo into new angles, new scenes and short clips.

Our take: one prompt did not remove the director; it moved the directing into 448 words. Writing a brief that precise, and spotting the lawn it missed, is the skill worth building now.

Source: Higgsfield on X, Nov 6, 2026; Luma, Nov 3, 2026.

Provenance — Higgsfield on X: 10-min doc one-shotted in Claude; Luma Labs on X: Luma MCP for Claude and ChatGPT · Status: verified: MEDIUM (two vendors, both self-promo; no independent check)
Higgsfield's one shot dinosaur film ran on a 448 word brief
Models

Gemini Omni 1.2 Flash makes a 15-second AI clip for $1.04, if you buy it from Google

Google made Gemini Omni 1.2 Flash generally available with 15-second clips, multi-turn editing and a 31% API price cut to about $0.07 a second at 720p. Runway had it live within the hour, but its help center still lists a rate about three times higher, 1080p and 4K are upscales, and every edit turn renders and bills a full new clip.

Drafted by The Generated · Filed by The Generated
Diptych: Google DeepMind chief executive Demis Hassabis in a suit, seated in conversation, beside Runway chief executive Cristobal Valenzuela speaking on a stage with a green backdrop

Google made Gemini Omni 1.2 Flash generally available on Thursday on the Gemini API (Google's paid service for developers) and Gemini for business. Clips now run up to 15 seconds, up from 10, you can edit a clip over several chat turns, and the API price drops 31%, to about $0.07 a second at 720p, or $1.04 for 15 seconds.

That makes Omni the cheapest major video model at 720p on list price this week, 26% under Kling 4.0 Flash, which opened to everyone a day earlier. But the number in the headlines is Google's paid developer rate. Runway had the model live within the hour, and its help center still documents a rate roughly three times higher. Editing a clip instead of re-rolling it isn't free either, because every edit turn renders and bills a whole new clip. Where you buy Omni matters more than the 30%.

Keep reading — 7 min

That’s all of Models.

6 Nov · 2 notes, filed as the record moved.

Models — archive

Also filed

AI made microdrama dubbing cheap, and the dubbing company's own books show who pays
AI dubbing made Chinese microdramas cheap to export, and Transn's own filing shows who pays

Transn says it localized 20,000+ Chinese microdramas with cloned voices, but its filing shows margins falling as AI cuts prices per order. Most titles were likely subtitled, and voice consent terms are unpublished.

China's top AI drama is a grassroots hit made entirely with ByteDance tools
A part-time crew made China's top AI drama, and every layer of it belongs to ByteDance

Into the Song has run ten 100-episode seasons on Hongguo since August. The viral debt story is wrong, the play counts belong to different counters, and the money figures are the team's own word.

Hyundai India is paying for a Diwali microdrama out of its social budget
Hyundai India is paying for a Diwali microdrama out of its social budget

Hyundai Motor India is funding a six-episode Diwali microdrama with Terribly Tiny Tales from its social budget and will judge it over 15 days. The budget and view counts are undisclosed, and nothing suggests AI was used.