Graphic novels and Storyboards

In this page we show some short stories composed by images. Unlike video, static images are today highly automatable at low cost. This has caused damage to the world of traditional illustration and comics, further reducing earnings and authorial protection. Although screenwriting/storyboarding schools, including the CSC, Centro Sperimentale di Cinematografia (fondazionecsc.it), continue to attract considerable demand and supply, the market has shifted. I am referring to small and mid-sized Italian film production companies — those with budgets around €0.5–2 million — where the use of storyboards has declined, without there actually having been a substantial replacement by automatically generated ones. This is caused not only by socio-political issues — distrust and exclusion — but also by the low quality of the off-the-shelf tools available on the web, with average subscriptions in the order of $35–50/month. To all this I would add a cultural factor tied to the asymmetry of the digital divide, which dominates sectors such as cinema — one that tends to be poorly suited to the times when it comes to pre-production tools and organisation/communication tools during the production phase, unlike the platforms on which Disney, Netflix, Amazon, Apple and so on operate. There, a business model geared towards cost optimisation and the remote deployment of resources favours meticulous attention to innovative technologies (both graphic and textual/communicative) across every production phase.

I have labelled the examples in this section: Storyboard and Graphic Novel. But strictly speaking, they are neither one nor the other. The stories in the Storyboard group lack the proper construction of bridging shots needed to evoke a more articulated visual narrative, useful as a support for directing. The graphic novels, meanwhile, are genuinely short. Here, rather than completeness, what I have tried to deliver is generative consistency, didactically preparatory to designing a true automated storyboard. Not by chance, each illustrated story linked above is accompanied by a Post (see link to ComfyUI section) with educational content that tries to describe the technology used. It is clear what the communication gap is today between what is being generated and how it is generated: the resulting images are seen as works of magic or sorcery one needs to defend oneself from, like the Sirens' song in the Odyssey. Let's take an example. YouTube tutorials on generative AI explain, in a practical and sometimes hasty way, the key steps for designing — or merely running — an image-generation process (video, sound, etc.) with a given tool. But on average they offer no explanation of how or why that technology works, what its extensibility might be, and in particular in which cases it produces poor results. I understand that the underlying formal models are not easy to popularise, but focusing only on the technical-manipulative aspect, and taking for granted a whole set of knowledge needed to understand and then try to replicate similar examples, means drifting conceptually away from the technology — thereby feeding a process of fear and helpless surrender to enslavement by it. This gap between going deeper and seeing an immediate result drives a proliferation of ready-to-use tools — the likes of Katalist, LTX Studio, Boords AI — presenting themselves as a magic hat, capable of doing everything: from ads to reels, from short stories to storyboards, with two prompts and a few keywords.

With that long preamble behind us, I focus on the primary objectives.

1. Verify the minimum input — i.e., the starting images — needed to generate a story that includes elements absent from the initial information. ComfyUI-based pipelines are now capable of covering a wide range of missing information without resorting to targeted training such as LoRAs and similar approaches. Although the use of custom training Nodes (such as LoRA specifically) on the story does allow for better visual quality of the story itself.
2. Maintain the identity of face, body, clothing and setting across a sequence of frames — this is the most critical weakness of off-the-shelf tools available online, which frequently lose the initial identity and visual consistency.
3. Preserve the coherence between the script text and the generated image — the image is not simply the product of prompt design, but the result of an orchestration of nodes, each of which is a neural network or a graphic function, controlled by a pipeline that forms its backbone, its frame (intuitively: bricks with input/output ports and segments connecting those ports).

In the next posts I intend to describe a more substantial and professional storyboard, starting from an external screenplay written by a professional friend: Riccardo Aprile, winner of a Solinas prize.