🔴Serious Discussion: Google Veo 3 and Flow -- AI Video Forgery Will Only Get Worse (Better)
(youtu.be)
BASED AI???
New here?
Create an account to submit posts, participate in discussions and chat with people.
You are viewing a single comment's thread. View all
Toast message
is a huge new industry, but limited to 11 second clips so far.
non-free non-local animation ?
- Pika 2.1 [https://pikartai.com/pika-2-1/]
- Kling [https://klingai.com/]
- Runway ml [https://runwayml.com/]
- Sora AI [ChatGPT Plus and ChatGPT Pro)]
those 4 ranked : https://www.youtube.com/watch?v=LWgwIU0XcrU
4 other non-free:
- https://ltx.studio/platform
- Invideo AI v3 - https://ai.invideo.io/ - https://www.youtube.com/watch?v=Qidf7v7UvVw
- Hailuo's Minimax - https://hailuoai.video/ $0.30 per 6s
- Google Veo 3 and Flow -- (best in May 2025) - https://youtu.be/pwFczfc0REU
= = = = = =
**Google Veo 3 and Flow** (for censored videos watermarked to your credit card)...
STEPS:
Create Google account
Get a USA credit card for funding Google AI Ultra ($249.00 per month) https://gemini.google/subscriptions/
Visit the official Google VEO 3 website.
Sign in with your Google account and join the VEO 3 Free AI trial.
Choose your prompt or import an idea from Flow Google.
Click “Generate” to see your video come to life in seconds.
8 second clips max
https://nofilmschool.com/veo-3-examples
wow.
= = = = = =
local:
Cogvideo in ComfyUI - CogvideoX 5B :
- https://stable-diffusion-art.com/cogvideox/
Stable Video Diffusion img2vid SVD
- Alibaba's Wan 2.1 ! **Mar 2025 open source FREE** - https://www.youtube.com/watch?v=0jSDmLtNe_k [ install tutorial ]
watch that info video
Wan Github: https://github.com/Wan-Video/Wan2.1
remote monitor :
CogVideoX Tora :
- https://www.runcomfy.com/comfyui-workflows/cogvideox-tora-image-to-video-model
free non-local :
- https://video.galaxy.ai/ai-animation-generator
- https://www.animon.ai/create
- Pollo AI - https://www.youtube.com/watch?v=ziVihe3kO5c
make some hoax comedy videos
In this case, they're pursuing Hollywood: [Lionsgate Inks Deal With AI Firm to Mine Its Massive Film and TV Library](https://www.hollywoodreporter.com/business/business-news/lionsgate-deal-ai-firm-runway-1236005554/)
Hollywood (especially advertising firms), and the "big ticket" clients that AI firms are pursing. As the video mentions, why hire actors, film and post-production crews for $250k when AI can make that 30-second film-quality TV ad for only $1,000? Agencies and ad-buyers are excited; now companies can make a new ad every week or every day, and they can make them quickly to capitalize on the news or latest meme.
But the biggest concern is how this tech can be exploited for evil. As one of the comments wrote, this has the power to change court rulings: *"Do you have evidence of this crime?" "Give me 5 mins. and I will."* There's definitely going to have to be federal laws over this tech; like making it a felony to make or publish any AI video, in any context, of any political candidate during an election.
No.
Each added second DOUBLES the needed VRAM!!!
I think 11 seconds is OVER 48 gigabytes needed for temporal coherency!
15 seconds is approaching half a terabyte.
= = = = =
workaround.... the last animation frame, if nothing in motion, can be used as start frame of a new 8 to 11 second clip.
= = = = =
A new invention might allow unlimited durations one day
That's what's happening with AI.
for coherence all prior data is used for each frame... worse... its not linear.
Making long, good-looking videos with video diffusion, especially using next-frame prediction models, is tricky due to two core challenges: forgetting and drifting.
Forgetting occurs when the model fails to maintain long-range temporal consistency, losing details from earlier in the video.
Drifting, also known as exposure bias, is the gradual degradation of visual quality as initial errors in one frame propagate and accumulate across subsequent frames.
11 sec uses a shitload of RAM
one new hack is "FramePack"
May 2025 FramePack:
======
FramePack is a Neural Network structure that introduces a novel anti-forgetting memory structure alongside sophisticated anti-drifting sampling methods to address the persistent challenges of forgetting and drifting in video synthesis. This combination provides a more robust and computationally tractable path towards high-quality, long-form video generation.
The central idea of FramePack’s approach to the forgetting problem is progressive compression of input frames based on their relative importance. The architecture ensures that the total transformer context length converges to a fixed upper bound, irrespective of the video’s duration. This pivotal feature allows the model to encode substantially more historical context without an escalating computational bottleneck, facilitating anti-forgetting directly.
FramePack system is built around Diffusion Transformers (DiTs) that generate a section of S unknown video frames, conditioned on T preceding input frames. It does not allow camera movement yet.
https://github.com/lllyasviel/FramePack
https://lllyasviel.github.io/frame_pack_gitpage/
TL/DR : XBX_X YOU ARE WRONG
RAM limits clip size
for coherence all prior data is used for each frame... worse... its not linear.
Making long, good-looking videos with video diffusion, especially using next-frame prediction models, is tricky due to two core challenges: forgetting and drifting.
Forgetting occurs when the model fails to maintain long-range temporal consistency, losing details from earlier in the video.
Drifting, also known as exposure bias, is the gradual degradation of visual quality as initial errors in one frame propagate and accumulate across subsequent frames.
11 sec uses a shitload of RAM
one new hack is "FramePack"
https://arxiv.org/pdf/2504.12626
May 2025 FramePack:
======
FramePack is a Neural Network structure that introduces a novel anti-forgetting memory structure alongside sophisticated anti-drifting sampling methods to address the persistent challenges of forgetting and drifting in video synthesis. This combination provides a more robust and computationally tractable path towards high-quality, long-form video generation.
The central idea of FramePack’s approach to the forgetting problem is progressive compression of input frames based on their relative importance. The architecture ensures that the total transformer context length converges to a fixed upper bound, irrespective of the video’s duration. This pivotal feature allows the model to encode substantially more historical context without an escalating computational bottleneck, facilitating anti-forgetting directly.
FramePack system is built around Diffusion Transformers (DiTs) that generate a section of S unknown video frames, conditioned on T preceding input frames. It does not allow camera movement yet.
https://github.com/lllyasviel/FramePack
https://lllyasviel.github.io/frame_pack_gitpage/
TL/DR : XBX_X YOU ARE WRONG
You are missing the whole point of why this tech exists. It exists to make forgeries of historical evidence, that is it.
It is now meaningless e.g. if you happen to have the video of the 9/11 towers with no plane/jet engine sounds. If you actually happened to have the 'Frazzledrip' vid, it now has no import as its provenance is sus. Not saying it was, but if the 'DE weapons caused the cali fires' actually was true, and you happened to have a video of it, it wouldn't matter. If bigfoot turned out to be real and whole families of them move into a town somewhere in the midwest, everyone would first suspect AI slop.
You will never be able to verify anything you didn't see yourself, all while pushing people to stay at home more and more. I know people who wouldn't leave their apartment unless you made them, and now they can collect a paycheck, get all of their food door-dashed, etc. One of these guys legit let his car battery die from not driving it. These guys already don't live in reality, their attack surface is so large already.
That would count as evil. And I mentioned how it should be illegal to publish AI videos involving candidates. There's already entire sites dedicated to prompt-based AI porn. Meaning someone or a group of people were able to train a model to make porn of women ... meaning it could easily do the same of children, right?
This is something that the DoJ and child welfare groups are seriously concerned about because current laws are specifically predicated, by definition and in their very language, on endangerment and exploitation. In other words, the content itself isn't illegal, because porn is 100% legal and even protected as speech; it's the fact that a child was exploited to make that particular content that makes it illegal. Well, does that mean no crime was committed if no human minor was ever exploited? It's literally just pixels, arranged in a particular way. This unforeseeable technology undermines so many laws, in so many fundamental ways. Any lawyer worth their salt would argue, "Where is the victim?"
Indeed, *"You may live to see man-made horrors beyond your comprehension."*