Faster images, cheaper video

Two releases this week moved the tools for brand content a step further. On 4 November OpenAI released Images 2.6, which makes animated GIFs and vector files and works faster than the version before. A day later Google released Gemini Omni 1.2 Flash, a video model that makes and edits longer clips for less money. Neither changes what a brand needs. Both change how quickly, and how cheaply, a team gets from an idea to something it can judge.

OpenAI: Images 2.6

Images 2.6 arrived in ChatGPT, Codex and the API, on every ChatGPT plan. OpenAI says it generates images 50 percent faster than before. In the API it comes as two models again: GPT Image 2.6 Flare, which at low quality “can generate images in as little as six seconds”, and Sunburst, for more detail and steadier results across edits. OpenAI also says Flare now lays out posters and product images better. The price per token has not changed since 2.5: 30 dollars per million image output tokens.

The visible novelty is motion. In ChatGPT a GIF can start from a prompt or a photo, and its speed and single frames can be adjusted afterwards, The Next Web reported. In the API a GIF has 2 to 16 frames, by default nine frames of 200 milliseconds each, a loop of 1.8 seconds. OpenAI suggests uses such as an animated brand mascot or a product shot turned into a loop.

For developers, image markup comes to the API, so an edit can be pointed at an exact spot, and a new complexity setting decides how much the model plans before it renders. Images can also come back as SVG files, which OpenAI says makes them ready to use on websites, in apps and in print. The documentation explains how: the API first makes a picture, then a second model draws it as vectors, and both steps are billed. The result is a static file, without scripts or embedded images.

Google: Gemini Omni 1.2 Flash

Google released Gemini Omni 1.2 Flash on 5 November and made it generally available in the Gemini API and to businesses through Google Cloud. Clips can now run up to 15 seconds, up from 10 with Omni 1.1, in any whole second from three. At 720p a second costs 0.07 dollars, 30 percent less than the 0.10 dollars of Omni 1.1, so a full clip of 15 seconds comes to about 1.04 dollars. Drafts at 360p cost 0.023 dollars a second, while upscales to 1080p and 4K cost more, up to 0.209 dollars a second.

The other changes are about control. A clip can be edited turn by turn, swapping a product or a background, changing the light or the aspect ratio, without generating it again. Google says faces, garments, packaging typography and brand colours stay consistent across camera angles, and the API takes up to six reference images. It also names its own limits: work continues on “fast sports action, rapid script changes and complex physics”.

Two details matter for teams in Europe. Editing or extending uploaded videos through the API is not available in the EEA, Switzerland or the United Kingdom, so a brand there cannot yet hand the model its own footage to rework; clips the model made itself can still be edited. And the old model leaves quickly: from 3 December, requests to Omni 1.1 are routed to 1.2 and billed at its rates.

What changes in production

First, finished formats come straight from the model. A looping mascot for a story, a product turning in a short loop, an icon as an editable vector: until now each meant a second tool and someone to convert the file. Now a first version comes out of the same request, and the designer starts from a working file rather than a flat picture.

Second, a second of video keeps getting cheaper. Ten drafts of a 15 second clip at 360p cost under 3.50 dollars, and only the chosen one needs an upscale. Variants for each audience and placement stop being a budget question, and Google’s own advice is to draft low, “fix a detail with a follow-up edit” and upscale only the best.

Third, rounds get shorter. A draft in six seconds and an edit that changes one thing instead of the whole clip mean more rounds in an afternoon. Both models also arrive where design teams already work: Figma Weave is quoted in both announcements, Adobe in Google’s, and Higgsfield is listed among OpenAI’s early access customers.

What stays the same

A faster model still does not know your brand. OpenAI’s documentation says the model “may occasionally struggle to maintain visual consistency for recurring characters or brand elements” across generations, the same caveat we noted in our review of Images 2.5 in October. Consistency still comes from a locked style: real references, fixed light, crop and colours, written down as a recipe and tested.

More output also means more choosing. Ten cheap variants are only useful if someone with taste picks the right one and can say why the others fail. In our view that job grows as generation gets cheaper, because the hard part moves from making to judging.

Rights and disclosure do not get simpler either. Both companies now watermark with the same system: OpenAI says it is adding Google DeepMind’s SynthID across ChatGPT, Codex and the API, next to C2PA metadata, and every Omni video carries SynthID. In the EU, the AI Act’s rules on marking and labelling AI-generated content have applied since 2 August 2026, with tools already on the market before then given until 2 December to mark their output, and the Commission’s code of practice describes how to meet them. A brand still needs the rights to every reference it uploads, and a clear rule on when it labels what it publishes.

And approval stays with a person. OpenAI’s system card is plain about GIFs: “Motion-related violations remain challenging to detect, and the safeguards may not catch all violations.” A looping mascot is published content like any other, and it needs the same check and the same sign-off before it goes out.

Where we stand

This is the work Reel and Tempo are built around. Reel, which is live, makes images and film in a brand’s locked styles, chooses the best model for each job with GPT Image and Gemini among them, checks every result and leaves the pick to a person. Tempo, also live, is our social media engine: its visuals come from Reel, and today a person approves every post. Faster and cheaper models make that process quicker. They do not replace the style, the pick or the approval.