A YouTube video can take days to make, then end up doing a hundred views a day in the back catalogue a few months later. Hardisk built Reflare around that problem: instead of making yet another video, try to make the existing ones work again.1 The official site describes the same loop: scan the back catalogue, generate thumbnail variants, then test them directly on YouTube.3

That matters because creator workflows are heavily biased toward the next upload. Much less attention goes to the hundreds of videos that have already been written, shot, edited and paid for.

Turning thumbnails into experiments

Reflare connects to a YouTube channel, scans the catalogue and suggests older videos to retest. In the launch demo, the user picks a video, asks for up to five new thumbnails and sets how long each one should run. The service generates the variants and rotates them automatically.1

Hardisk demonstrates it on a 2023 video that had reached close to a million views but had fallen below roughly one hundred daily views. Reflare analyses the topic, the channel's visual language and, according to the presentation, YouTube data around similar content before producing new options.1

YouTube's own documentation adds an important limit. Changing a title or thumbnail can change performance because viewers react differently to the new packaging. The act of changing it does not itself trigger a reranking.2

So this is not an algorithm wake-up spell. It is a way to put a different proposition in front of viewers and measure what happens.

A catalogue is still material

This is the most transferable part of the idea.

Music, film and publishing already treat old catalogues as assets that can keep circulating. YouTube teams often spend most of their energy on the next release. Reflare applies a maintenance loop to the archive instead: observe, repackage, test and keep the version that performs better.

The product goes beyond image generation. It automates the test cycle and presents an impact estimate. Hardisk says the team worked with Google on a probabilistic method intended to separate traffic caused by the tests from views that would have happened anyway.1

That is still a vendor claim. There is not enough public technical detail or independent benchmarking yet to validate the quality of that attribution.

The same caution applies to the beta testimonials in the launch. Several creators say they have used Reflare for months and describe time savings or performance improvements. They show a plausible workflow, not an independent average result.1

The AI disappears behind the button

One small detail in the demo says a lot. Hardisk points out that there is no chat box. The creator chooses a video and starts a test. The AI sits inside the workflow rather than becoming the workflow.1

That is more useful than attaching an “Ask AI” field to every piece of software. The actual job is to find a fading video, create visual hypotheses, run them for long enough and decide what to keep.

Reflare tries to absorb prompting into the product. The model becomes one production step among several.

The service was presented as available at launch, with plans starting at €49 per month.1 The more interesting shift is economic: the creator is not only paying to package the next upload, but to maintain an existing body of work.

There is still a tension worth watching. Continuous optimisation may help people rediscover good old videos. It can also turn an archive into a product that is never allowed to stop being marketed.