Seedance 2.0 and Animation
AI animation is much further along than realistic cinematography
As I’ve continued to experiment with Seedance 2.0, I’ve discovered that it’s much easier to work on using it to animate an existing comic than to create new material from scratch. The reason boils down to model collapse: the system can only handle so many clips beginning from a frame from the previous clip before the character begins to visibly deteriorate.
Compare, for example, the face of the same character from one of her first appearances to her face in the final one less than two minutes later:
She looks as if she’s aged 30 years from the one clip to the next. This is because AI cannot accurately replicate an AI image; it always needs new information from which to work. Animating from an existing comic works very well because every frame is essentially an animatable storyboard and provides the AI with a new starting point with which to begin. The amount of model collapse that takes place over the course of 2-3 clips is minimal, which allows the creator to get 2-4 clips out of a single frame.
Below is a one-minute video of the animated THE TRAGEDY OF THE TRIBUNE, which is a comic based on the novel A THRONE OF BONES. While there are still some minor continuity issues, they are much less significant than in the two-minute THE GHOSTS OF BANGKOK and are of the sort that could be easily edited out by removing a few frames.
Note that as before, the video below is completely unedited in any way, it’s just a series of 14 5-second clips stitched together with no video or audio modifications. I would say that, in theory, it is possible to produce a professional-level animation now from existing comics, as opposed to my 18-24 month estimate before realistic cinematography will be viable.
The animated segment is already at the broadcast or theater quality of 50s-60s Hanna-Barbera or Warner Bros. cartoons. The steps to a longer feature look easier. Some crossover art to bridge scenes would make MP4 jumps easier to cover - another feature of older cartoons.
There we go, it says a lot of how viable it is with that minimal edit.
Interesting exciting times.
Huge leap for Arkhaven if you get ahead of the curve.
Incidentally it's of limited use due to not knowing the particular methods used, but recently was made aware of Gossip Goblin's The Patchwright, a 20+ minute AI work represents one of the more robust examples we have of how far the machine can be pushed. The particular style and the crowded detail cyber/fleshpunk aesthetic cover a certain amount of the inconsistencies, a part further by the narrative's setting. When everyone is modded to the gills, what even is visual information?
Though, notably at least the major characters are voice acted by humans.
An interesting question is who exactly his investors are, but there's undoubtedly quite an amount of people with an interest in seeing this tech bear fruit.
If anything it's surprising there isn't more people and their works making the rounds and getting snapped up by big money. Presumably there's a certain amount going on behind the scenes, but it has to be admitted the rapid pace has made for unstable ground for long form projects.
That said, this teller isn't familiar with the growing scene.
Only other significant wranglers he's aware of are Skybrowes (music videos) and whoever does the Archive Inbetween stuff.
Neuralvids was another one.
That's all the osmosis has gathered.
May your projects go smoothly gain unexpected hands and reap a fine harvest...