EDITING

The following are edit tests done with regard to using AI methods for dramatic storytelling. In an ecosystem where many are striving to tackle VFX and social media content, I have gone from thinking of the merits of previsualisation, to circa IO 2025 / WAN 2.2 feeling that actual medium form content could be achieved. Below are a series of tests and WIPs spanning October 2025-May 2026 demonstrating the growing ambition of this approach.

Everyones making memes, or doing i2v on playblasts and preview renders.

I've had my eye on drama. Been trying to take a tv production mindset to an ai inference game, via animation bias. In total, the showcase below features over 10 minutes of purely AI-generated content.

Techstack: ComfyUI, Musubi Tuner, AI Toolkit, Adobe Premiere, DG_Directorview (proprietary), Autodesk ShotGrid, Google Sheets

Models: LTX-2.3, LTX-2.0, WAN 2.2

Githubs: POM/desloppify, alvdansen/klippbok

Glenroe

TV EPISODE WIP / LTX.23 / COMFY UI / MUSUBI TUNER

Following on from training LoRAs on low fidelity images, I was inspired to wonder how-low-can-i-go. Glenroe, a rural Irish soap opera broadcast from the 1980s-2000, is all but forgotten today save 15 or so VHS uploads from tapes. State broadcaster RTE notoriously has lost all broadcasts prior to 1977. Coupled with the closed archive I felt that resurrecting an iconic, you had to have been there at the time, television program would work as a technical exercise, nostalgic trip down memory lane, and a perfect benchmark for my intent of using AI for drama, rather than action or memes. Very much a WIP, work on DG Glenroe was disrupted by a necessary shift from a ShotGrid database to a custom one. Characters doubling, LoRA context bleed between characters, and refinement of prompting methodology is evident in the most recent shots done with my advanced prompting logic. I have yet to match to the higher fidelity tests that are crafted from a single prompt block.

Conors Deadly Demo

YOUTUBE TUTORIAL WIP / LTX 2.3 / COMFY UI / MUSUBI TUNER

I hate hearing my own voice on speakers and so when I started to prepare for documenting my work, I wondered how to mix staying truthful to my Dublin heritage. Who would be a more boisterous Dub who could narrate and guide users through my test suite?

UG Police

MUSIC VIDEO WIP / WAN 2.2 / COMFY UI / PREMIERE

A longstanding personal project featuring abstracted characters. Generated in wan2.2 using comfyui, this was a narrative-less montage of a day in the life of a primate looking security guard. Like all entries on this page, the short music video was generated in a 'can this LoRA i've just trained actually work for storytelling?, lets spend a Sunday afternoon editing in Premiere'. Like all on this page its an aging WIP from which notes have been taken, but the project has not been iterated. True to all of my projects the proof of concepts lead to vibecoding development work so that they benefit those who come after. If not for job applications this would remain offline.

ULEZ

MUSIC VIDEO / WAN 2.2 / COMFY UI / AI-TOOLKIT

In the United Kingdom the ULEZ initiative, ultra low emission zones, have flourished to the chagrin of selected motorists who have named themselves the Blade Runners. These brave heroes drive out in the dead of night to find ULEZ cameras atop traffic lights near schools, and fell the whole traffic light, prohibiting safety, and brandishing their achievements with pride as they destroy public property. Against the backdrop of a Prime Minister who regularly tries to seduce the alt-right vote, this is a short narrative proof of concept, moving the WAN 2.2 work into COMFYUI.

PHARAGE

MUSIC VIDEO / AI-TOOLKIT

A slow network on vast.ai left me with a 4hr download for my LoRA training checkpoint, so I asked the computer to generate me 7 minutes of retro dystopian nightmare fuel featuring alt-right agitator Nigel Farage. I genuinely did not realise before this that prompt stacking could lead to narrative driven storytelling, most of the throughput being memes or hallucinatory VFX. All footage generated in AI-TOOLKIT.

← Back to Studio