Rising up, if I needed to experiment with one thing technical, my dad made it occur. We shared dozens of tech adventures collectively, however these adventures have been reduce quick when he died of most cancers in 2013. Because of a brand new AI picture generator, it seems that my dad and I nonetheless have yet one more journey to go.
Just lately, an nameless AI hobbyist found that a picture synthesis mannequin referred to as Flux can reproduce somebody’s handwriting very precisely if specifically educated to take action. I made a decision to experiment with the approach utilizing written journals my dad left behind. The outcomes astounded me and raised deep questions on ethics, the authenticity of media artifacts, and the private which means behind handwriting itself.
Past that, I am additionally comfortable that I get to see my dad’s handwriting once more. Captured by a neural community, a part of him will dwell on in a dynamic approach that was not possible a decade in the past. It has been some time since he died, and I’m not grieving. From my perspective, it is a celebration of one thing nice about my dad—reviving the distinct approach he wrote and what that conveys about who he was.
I admit that copying somebody’s handwriting so convincingly might deliver risks. I have been warning for years about an upcoming era the place digital media creation and mimicry is totally and effortlessly fluid, however it’s nonetheless wild to see one thing that appears like magic work for the primary time. It is tempting to say we’re getting into a brand new world the place all types of media can’t be trusted, however in actual fact, we’re being given additional proof of what was at all times the case: Recorded media has no intrinsic truthfulness, and we have at all times judged the credibility of data from the status of the messenger.
This fluidity in media creation is completely exemplified by Flux’s strategy to handwriting synthesis. One of the crucial fascinating issues in regards to the Flux resolution is that the ensuing handwriting is dynamic. For essentially the most half, no two letters are rendered in precisely the identical approach. A neural community just like the one which drives Flux is a big net of chances and approximations, so the imperfect circulate of handwriting is a perfect match. Additionally, in contrast to a font in a phrase processor, you’ll be able to natively insert the handwriting into AI-generated scenes, equivalent to indicators, cartoons, billboards, chalkboards, TV photos, and rather more.
It is value noting that neither I nor the one that lately found that Flux can reproduce penmanship have been the primary to make use of neural networks to clone handwriting—analysis into that extends again years—however it has lately turn out to be virtually trivially cheap to take action utilizing both a cloud service or consumer-level {hardware} when you’ve got the writing samples available.
Here is how I introduced a chunk of my dad again to life.
The invention
As a every day tech information author, I regulate the newest improvements in AI picture technology. Late final month whereas looking Reddit, I seen a submit from an AI imagery hobbyist who goes by the title “fofr“—pronounced “Foffer,” he instructed me, so let’s name him that for comfort. Foffer introduced that he had replicated J.R.R. Tolkien’s handwriting utilizing scans present in archives on-line.
Foffer initially made the Tolkien mannequin accessible for others to make use of, however he voluntarily took it down two days later when he started to fret about individuals misusing it to create handwriting within the model of J.R.R. Tolkien. However the handwriting-cloning approach he found was now public information.