A big step since the days of negative cutting 16mm film...
A few weeks back we received a limited offer to try out for free an AI video generator, VideoAI. Text to video. Sounded simple enough. (On-line) ya' type in a basic but graphic description of the shot or short scene ya' want to produce (a "prompt" in the vernacular), hit the "generate" button, wait a few minutes, then voilà, you're presented with an AI generated rendering of your concept - character, set/location, audio effects - the works. Ok. In our trial run we kept it simple enough - "twilight, a flying saucer landing in a rural cow pasture, the flying saucer emitting a high pitched humming sound, the sound of a cow bellowing" - pretty much our prompt word-for-word. Was mighty impressed with the result.
For the past year we'd been working on and off (when inspired) on a new stop motion short. Part experimental, part venture, always fun. We'd adapted a brief script from an old comic book anthology. We'd had most of the props constructed. We'd photographed a few test shots. Out of curiosity we did a second VideoAI trial using a shot description, pretty much word-for-word, from a scene within our script. Again, was more than impressed with the results.
(©Marvel Comics)
The four-page concept inspiring 'Monsteroso'. A Steve Ditko ©Marvel Comics tale first published in 'Tales To Astonish' ©Marvel Comics magazine dated April, 1961. If the above disappears ya' know why. Otherwise, thanks to ©Marvel Comics for use of the images.
Stop motion creature prop.
A few (very rough) stop motion tests
From script, prompt for initial VideoAI test of creature
We'll leave reveal of the AI generated creature for the video viewing below
So impressed that we decided to abandon the stop motion premise and (further attempt to) complete the short utilizing VideoAI. Not without a few anticipated hitches, however. The main hitch being that each generation (shot) would be stand alone. Meaning that there is no continuity between generations. Matching shots, and primarily the creature shot-to-shot, could be a major problem. Second (but not really) was that generations maxed at a total length of eight seconds. We hint at not really, as we very seldom extend a shot beyond an average of six seconds. We could (typically) extend any longer shots with a transition insert. In turn, that sounded like fun, harkening back to the old analog film days of maybe 17 seconds of footage on a spring-wind camera. Would take a bit more planning with concern to maximum shooting ratio (under 1.25:1) and budget, as each generation costs $5. Peanuts for a short, under ten minute film. But a bucks a buck. And PBR's are currently up to $25/case.
The completed video:
Nano X (2026)
(Running Time: 5:36)
Comments:
AI Generator: Google Veo 3
A final shooting ratio of around 1.2:1
Not too bad. We figured out a few tricks to (somewhat) maintain continuity. Re-generating a shot for new and re-editing the prompt while maintaining the creature description, or cutting/pasting a prompt to a new generation then re-editing seemed to work well for keeping the creature consistent. Any resulting glitches were minor. There were some noticeable differences with the coverage of the creatures fur here and there. One close shot included black pupils in specified solid yellow irises. We re-edited subsequent prompts to specify pupil-less. One generation added two large saber-tooth fangs to the creature, without prompting that addition. Was pretty cool looking but up to that point it was too late to change its appearance. We worked around it by clipping the shot. And we're not crazy about the un-prompted huffs of AI added dust-like breath. And a few times the generator added a music cue which was a somewhat minor annoyance in post-editing as the cue and sound effects are single track.
Figuring that (at least Google) AI is this unified consciousness, another trick was to compose a prompt then paste it into the Google search bar, asking Google AI to describe the described. We would then re-edit our prompt if and/or where we felt it necessary to satisfy AI's interpretation.
Prompting shots as wide angle gave us the added option of scaling and cropping the shots in post-edit for re-use as alternate perspective close to medium shots or transitions, tripling the use of a generation or two.
Other than the end shot AI generated "microbe", the entire final "lab" sequence was all physical props and computer-keyed background plates. The "Test Results Interpretation" are six Mac/Pages software generated screenshots scaled and cropped in post-edit.
Music and sound FX obtained from Pond 5.
All told the current AI generated text-to-video technology is pretty cool. Although we don't follow the trend, it's a bet that we probably experienced the absolute basics of what has been so far developed. If not already existent, we're sure that rapid improvements, from generation to generation continuity, to being able to interact with the AI generator just as a director interacts with a production crew (minus the crew and expense) are forthcoming. Actors will exist in computer files. We can see why Hollywood would be a bit nervous. Envision an industry of only screenwriters and digital filmmakers and designers and IT engineers and technicians.
One thing pretty cool will be the ability to seamlessly bring back actors of old. A few films we'd like to see:
How about John Wayne battling Predator in the late 1800's American southwest. Someone mentioned to us that they read about that a while back. Sounds as stupid and ridiculous as it does fun.
Arnold Schwarzenegger in the Twelve Labors Of Hercules, as dictated by Apollodorus. We can envision Arnie on a mountaintop high over Lake Stymphalia chucking arrows at swooping flocks of metallic feathered, man eating Stymphalian birds. Gotta' be Arnold, though, or we ain't watchin'.
A Sean Connery Bond remake of Dr. No and Moonraker. Both as literal novel adaptations. Would look forward to the ventilation system obstacle course (with spiders) as well as the giant squid encounter of Dr. No. And, Honey Rider sans bikini.
Disclaimer to the above: Informative content for on-line/private viewing only. No monetary compensation is involved.
No comments:
Post a Comment