A user on Bilibili uploaded a Jujutsu Kaisen animation created using Seedance at extremely low cost. I know many people on the forum have shared even more impressive AI videos, but I never imagined it could resemble the original anime so closely.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
A user on Bilibili uploaded a Jujutsu Kaisen animation created using Seedance at extremely low cost. I know many people on the forum have shared even more impressive AI videos, but I never imagined it could resemble the original anime so closely.
Here are some even better Jujutsu Kaisen AI animations. However, the creator mentioned that Seedance's content review has become extremely strict, severely impacting production. For instance, copyrighted characters and real-person images are now prohibited, while sensitive terms are subject to stricter restrictions. As a result, new Jujutsu Kaisen videos won't be released anytime soon. Some also claim Seedance's generation quality has significantly declined since its initial launch. To make matters worse, even with a premium membership, generating a single video now requires over a five-hour wait in the queue.
I believe we are on the eve of an AI creative explosion. While AI currently faces numerous regulatory and computational constraints, these will gradually resolve over time. With just a final push of technical optimization, we'll enter an era where anyone can affordably create videos that once required high costs. I anticipate that sometime next year, the internet will witness a technological leap akin to the transition from 3G to 4G networks, leading to an explosion of AI animation content creators.
broadsword
Brigadier
With Seedance, you can skip motion capture, animators, etc. It saves the studio a lot of money. Hollywood does not care how you do it as long as the end product is good.
AI-generated motion and models are also maturing. Future workflows will likely involve setting up models and shots in UE5, shooting drafts as a reference for the video generation model
Can AI currently generate motion data that can be directly imported into animation software? It would be ideal if AI could directly generate motion data for Blender models, as this would significantly improve scene stability compared to AI generating videos directly.
The architecture itself is the fundamental bottleneck. "AGI" is supposed to be able to tackle out-of-distribution problems by reasoning. This is impossible, pretty much by definition, for a statistical model. It's only modeling the distribution for the training data. Just to be clear, there is no way the smartest people in this industry don't already know this, it's just the US doesn't have the ability to rein in Silicon Valley's reckless expenditures.
AGI might come someday, but ACI ("artificial capable intelligence", i.e. AI smart enough to replace 90% of entry-level office/tech workers) is definitely coming. The funny thing is once companies replace junior positions with AI, they're eventually going to have a difficult time filling senior roles as boomers retire or switch jobs and they don't have any young'uns working in junior roles to get promoted.
Tencent Hunyuan Motion 1.0 can generate some basic actions. For now, video motion capture is a more feasible approach, such as GENMO, GVHMR, Easymocap etc.
Doesn't matter the ranking. People vote with their feet.
Current size of China's Enterprise AI market. Qwen is in the lead, followed by Doubao and DeepSeek. Of course, this was done before GLM-5, Kimi 2.5 & Minimax 2.5 came out, which are driving a lot of token generation.