Futureprint

View Original

FuturePrint Tech AI for print conference review

By Michael Walker

Narrow AI. Machine learning. Deep learning. Industry 5.0. Ground truth. Adversarial networks and convolutional neural networks. Natural language processing. Just like print, artificial intelligence (AI) has its own jargon, technologies and capabilities. To get a better understanding of what those are and how specifically they apply to the print industry, FuturePrint Tech held the AI for Print conference in Cambridge in April 2024.

The event brought together some 15 speakers from a wide range of backgrounds and covered a lot of ground. Some, such as Finn Karsten of KnowledgeBrief, provided an overview of AI developments and made the distinction between existing ‘narrow’ single-purpose AI, like content or product recommendation engines, and ‘true’, more broadly applicable AI that can adapt to new tasks in the way that humans can. He gave examples of current applications in healthcare and education, but existing and anticipated AI applications relevant to print broadly fall into four categories – content creation, prepress, production and process optimisation, and quality control.

The creative applications span text, image, texture/pattern and layout creation. Some of the speakers used ChatGPT, for example, to help produce first drafts, content suggestions and to research ideas for their presentations. While they thought that it was a powerful tool to get things started, all agreed that the results needed evaluation and further development by humans, whether that meant further rounds of directed AI iteration, or fully ‘biological’ processing.

AI image creation examples were also presented by several speakers, both to illustrate generative AI’s evolving abilities, and how it can also go awry, confirming that the software doesn’t actually ‘understand’ the images it’s creating. A distinct AI ‘style’ of illustration also appeared across several presentations, with humanoid robots and translucent information screens with blue displays appearing in several.

Other applications that aren’t print-specific include personalisation, content recommendations, translation / localisation services and customer services, with chat bots as a frequently-found example. Applications in education were also discussed by Elizabeth Bowerman of specialist printer Stephen Austin & Sons.

Image issues

Within prepress, some AI application address pre-flighting issues, such as processing images of insufficient resolution for the planned output size and method, for which there are already tools to intelligently increase image resolution. A related issue is generating missing image bleed, for which AI-based tools are also available now. An interesting example was shown by Matt Dass of label specialist Springfield Solutions, who had used Adobe’s Firefly generative AI to create a spot varnish layer for a for moisturiser jar label, based on the image in the CMYK artwork.

But before anything goes into prepress, you need an artwork file. Alan Bendall of DX Consulting described a digital asset location plug-in/API tool that allows branded artwork to be searched visually by AI rather than relying on file metadata being manually searched by operators. He claimed that in one customer instance, finding the right files and determining the optimal output path (printer location, press and substrate type) had been reduced from 55 minutes to under one minute.

Another application that’s under development at HP is in colour management, including rapid press profiling using minimal ink, achieved by focusing on the colours that show the greatest deviation. HP’s Jan Morovic explained how these profiling techniques can be used for both colour correction and for aesthetic or design purposes. He also described how AI can be used for cross-substrate colour prediction in digital printing of textiles, where matching brand or team colours across a range of fabrics is important.

Component optimisation applies in printhead design, where Simon Jelley of 42 Technology explained how trial-and-error testing of control waveforms can be cut short by AI-suggested solutions. There are potential AI uses too in making automated choices about screening techniques in Rips, depending on image content and print type. Together with Ripping time prediction, something that David Stevenson of Global Graphics Software showed has already been patented, this would allow for reliable and efficient machine assignment decisions to be made; further developments in a similar vein would allow for both supply chain and business process optimisation.

We are already seeing AI use in pre-emptive equipment maintenance and failure prediction, and real-time quality control is another example given by Simon Jelley. Linked to computer vision systems, AI software can be taught recognise print defects, check for both colour and registration accuracy, and eventually perhaps even suggest reasons for the problems it detects.

Having the right problems

More than one speaker also pointed out that AI isn’t a magic bullet and that there are various cases where a well-tuned conventional algorithm will do a better job, such as in nesting/ganging and imposition. Simon Jelley spent part of his presentation describing what types of challenges AI is suited to: problems that are too complex for an algorithmic approach; where there is no one ‘right’ answer; where there is plentiful data available for training the AI; and good ‘ground truth’ data with known classification or scoring to help the AI.

Speaking from experience gained in the development of the SmartDFE technology at Global Graphics Software, David Stevenson noted that AI can ‘throw up strange predictions, highlight missed nuances or allow experimentation with different approaches’. The chances are that one way or another most printers will wind up using AI in their operations, but the overwhelming advice is to proactively seize the opportunity.

According to Royce Dodds of Dekor8 and the Digital Printing Association, ‘AI is just about leaving the platform – if you could be getting on board, you should be’.

HP’s Jan Morovic put it more strongly, saying ‘AI will not replace PSPs [print service providers]; PSPs using AI will replace PSPs who don’t’.

Michael Walker was editor of Digital Printer magazine from 2017–2024 and has been writing about graphic, prepress and print since 1987.