Today's brief: more tools within reach

This edition covers three updates announced on 13 October and an official ChatGPT Space demonstration published on 14 October. They tackle different parts of the same working day: building something complicated, changing a filmed character, teaching a video model a particular look, and keeping a team's decisions together.

These are source-verified updates, not XANTRANS hands-on reviews. The examples below are suggested uses. A polished launch demonstration does not establish how reliably a tool will handle your own footage, code or project.

Gemini 4 Argon: the wider rollout has begun

Google's 13 October announcement moves Argon beyond its restricted September preview. Developers can use the paid Gemini API and AI Studio. Antigravity access is for Google AI Ultra subscribers, while the Gemini app and AI Mode include Pro or Ultra routes. Those are different products: a consumer Pro subscription does not mean Pro-tier API access. Availability still needs checking in the particular account and product.

For a small studio, a useful trial would be a defined website change across several files, with acceptance criteria and tests supplied up front. Ask for a reviewable result, then inspect it. Google's examples suggest longer coding and analysis jobs, but they are provider demonstrations, not a guarantee that a project can run unattended.

The API has a shared 1,048,576-token window covering input, thinking and output. It is not a million output tokens on top of a million input tokens. Normal synchronous output defaults to a 65,536-token maximum; longer output requires background execution and uses what remains of that shared window.

Introductory API pricing runs through 30 November: $2 per million input tokens and $10 per million output tokens, including thinking. From 1 December, the documented rates become $4 and $20. These are usage charges, not app subscription prices. A longer run can also take longer to inspect, so the first test should be small enough to judge.

ElevenCreative Recast: change the character in an existing shot

ElevenLabs introduced Recast in an official 13 October video, linking to an ElevenCreative template. Its public description promises replacement of the main character while retaining the source motion, scene and camera. That makes it a packaged creative workflow, not evidence of a newly released foundation model.

One practical use is to perform a short scene once, then explore a fictional character's appearance while keeping the performance as the starting point. For an advert or storyboard, this could help compare character directions without restaging every version. Check hands, contact with props, facial movement and the start and end of the shot before choosing a result.

ElevenCreative Templates are in beta. Browsing includes the Free plan, but video generation requires a paid subscription and consumes credits. The cost depends on the models in the template and can be previewed before running. The public material checked here does not establish Recast's exact duration or resolution limits, a fixed price, or guaranteed soundtrack and lip-sync preservation.

LTX Trainer: custom characters and styles inside Explore

LTX's 13 October update adds a hosted route to Character and Style LoRA training in Explore. The trainer itself is older; the change is being able to use it without setting up a local GPU training pipeline. A LoRA is a smaller adaptation that teaches an existing model a particular visual subject or style.

The workflow asks for a name, a trigger word and at least five minutes of footage across the uploaded videos. It checks clips for issues such as text, extra people and inconsistent appearance. Flagged clips are excluded by default, although the user can restore them. LTX describes a background process lasting several hours, with an email when it is ready.

A possible studio use is a recurring fictional character across several short scenes. The useful work starts with consistent, suitable source footage. More clips do not automatically mean better training if they introduce contradictory looks or distracting material.

The result can be tried in Explore and downloaded as weights. LTX says it remains available on the platform for 30 days, so download it if you need to retain it. No local GPU requirement does not mean no cost: the tutorial includes a billing-key step, and the written release checked here does not establish a fixed training price. Confirm billing and usage terms before starting a run.

Also useful: a shared project page in ChatGPT Space

OpenAI's 14 October demo shows a conversation becoming a shared working page. For a client project, the useful application is straightforward: keep the agreed brief, decisions and next actions in one editable place, instead of making everyone reconstruct them from a chat history. This is a workflow example, not a claim that Space launched today.

Current help documentation lists creation and editing for Pro, Business and Enterprise on web and desktop; mobile supports reading and sharing. A shared Page does not expose the underlying private chat or memory, but anything placed in the Page becomes visible to its collaborators.

One boundary matters: the written help still says automatic Keep updated is unavailable at launch, with Sheets and Slides coming later. Treat a maintained project page as the available workflow, rather than promising an automatically synchronised reporting system.

Sources & verification

Prepared from dated maker announcements, product documentation and official demonstrations. XANTRANS has not run these tools for this article. Suggested uses are editorial examples, not tested results.