The Rise of Hunyuan Video Deepfakes

Because of the nature of a few of the materials mentioned right here, this text will comprise fewer reference hyperlinks and illustrations than typical.

One thing noteworthy is at the moment occurring within the AI synthesis group, although its significance might take some time to change into clear. Hobbyists are coaching generative AI video fashions to breed the likenesses of individuals, utilizing video-based LoRAs on Tencent’s lately launched open supply Hunyuan Video framework.*

Click on to play. Various outcomes from Hunyuan-based LoRA customizations freely accessible on the Civit group. By coaching low-rank adaptation fashions (LoRAs), points with temporal stability, which have plagued AI video era for 2 years, are considerably decreased. Sources: civit.ai

Within the video proven above, the likenesses of actresses Natalie Portman, Christina Hendricks and Scarlett Johansson, along with tech chief Elon Musk, have been skilled into comparatively small add-on recordsdata for the Hunyuan generative video system, which could be put in with out content material filters (similar to NSFW filters) on a consumer’s laptop.

The creator of the Christina Hendricks LoRA proven above states that solely 16 pictures from the Mad Males TV present had been wanted to develop the mannequin (which is a mere 307mb obtain); a number of posts from the Steady Diffusion group at Reddit and Discord affirm that LoRAs of this type don’t require excessive quantities of coaching information, or excessive coaching instances, usually.

Click to play. Arnold Schwarzenegger is dropped at life in a Hunyuan video LoRA that may be downloaded at Civit. See https://www.youtube.com/watch?v=1D7B9g9rY68 for additional Arnie examples, from AI fanatic Bob Doyle.

Hunyuan LoRAs could be skilled on both static pictures or movies, although coaching on movies requires better {hardware} assets and elevated coaching time.

The Hunyuan Video mannequin options 13 billion parameters, exceeding Sora’s 12 billion parameters, and much exceeding the less-capable Hunyuan-DiT mannequin launched to open supply in summer season of 2024, which has only one.5 billion parameters.

As was the case two and a half years in the past with Steady Diffusion and LoRA (see examples of Steady Diffusion 1.5’s ‘native’ celebrities here), the foundation model in question has a far more limited understanding of celebrity personalities, compared to the level of fidelity that can be obtained through ‘ID-injected’ LoRA implementations.

Effectively, a customized, personality-focused LoRA gets a ‘free ride’ on the significant synthesis capabilities of the base Hunyuan model, offering a notably more effective human synthesis than can be obtained either by 2017-era autoencoder deepfakes or by attempting to add movement to static images via systems such as the feted LivePortrait.

All the LoRAs depicted here can be downloaded freely from the highly popular Civit community, while the more abundant number of older custom-made ‘static-image’ LoRAs can also potentially create ‘seed’ images for the video creation process (i.e., image-to-video, a pending release for Hunyuan Video, though workarounds are possible, for the moment).

Click to play. Above, samples from a ‘static’ Flux LoRA; below, examples from a Hunyuan video LoRA featuring musician Taylor Swift. Both of these LoRAs are freely available at the Civit community.

As I write, the Civit website offers 128 search results for ‘Hunyuan’*. Nearly all of these are in some way NSFW models; 22 depict celebrities; 18 are designed to facilitate the generation of hardcore pornography; and only seven of them depict men rather than women.

So What’s New?

Due to the evolving nature of the term deepfake, and limited public understanding of the (quite severe) limitations of AI human video synthesis frameworks to date, the significance of the Hunyuan LoRA is not easy to understand for a person casually following the generative AI scene. Let’s review some of the key differences between Hunyuan LoRAs and prior approaches to identity-based AI video generation.

1: Unfettered Local Installation

The most important aspect of Hunyuan Video is the fact that it can be downloaded locally, and that it puts a very powerful and uncensored AI video generation system in the hands of the casual user, as well as the VFX community (to the extent that licenses may allow across geographical regions).

The last time this happened was the advent of the release to open source of the Stability.ai Stable Diffusion model in the summer of 2022. At that time, OpenAI’s DALL-E2 had captured the public imagination, though DALLE-2 was a paid service with notable restrictions (which grew over time).

When Stable Diffusion became available, and Low-Rank Adaptation then made it possible to generate images of the identity of any person (celebrity or not), the huge locus of developer and consumer interest helped Stable Diffusion to eclipse the popularity of DALLE-2; though the latter was a more capable system out-of-the-box, its censorship routines were seen as onerous by many of its users, and customization was not possible.

Arguably, the same scenario now applies between Sora and Hunyuan – or, more accurately, between Sora-grade proprietary generative video systems, and open source rivals, of which Hunyuan is the first – but probably not the last (here, consider that Flux would eventually gain significant ground on Stable Diffusion).

Users who wish to create Hunyuan LoRA output, but who lack effectively beefy equipment, can, as ever, offload the GPU aspect of training to online compute services such as RunPod. This is not the same as creating AI videos at platforms such as Kaiber or Kling, since there is no semantic or image-based filtering (censoring) entailed in renting an online GPU to support an otherwise local workflow.

2: No Need for ‘Host’ Movies and Excessive Effort

When deepfakes burst onto the scene on the finish of 2017, the anonymously-posted code would evolve into the mainstream forks DeepFaceLab and FaceSwap (in addition to the DeepFaceLive real-time deepfaking system).

This technique required the painstaking curation of hundreds of face pictures of every identification to be swapped; the much less effort put into this stage, the much less efficient the mannequin could be. Moreover, coaching instances different between 2-14 days, relying on accessible {hardware}, stressing even succesful programs in the long run.

When the mannequin was lastly prepared, it might solely impose faces into current video, and often wanted a ‘goal’ (i.e., actual) identification that was shut in look to the superimposed identification.

Extra lately, ROOP, LivePortrait and quite a few comparable frameworks have offered comparable performance with far much less effort, and infrequently with superior outcomes – however with no capability to generate correct full-body deepfakes – or any ingredient aside from faces.

In contrast, Hunyuan LoRAs (and the same programs that can inevitably observe) enable for unfettered creation of complete worlds, together with full-body simulation of the user-trained LoRA identification.

3: Massively Improved Temporal Consistency

Temporal consistency has been the Holy Grail of diffusion video for a number of years now. Using a LoRA, along with apposite prompts, provides a Hunyuan video era a continuing identification reference to stick to. In idea (these are early days), one might prepare a number of LoRAs of a selected identification, every carrying particular clothes.

Underneath these auspices, the clothes too is much less more likely to ‘mutate’ all through the course of a video era (for the reason that generative system bases the subsequent body on a really restricted window of prior frames).

(Alternatively, as with image-based LoRA programs, one can merely apply a number of LoRAs, similar to identification + costume LoRAs, to a single video era)

4: Entry to the ‘Human Experiment’

As I lately noticed, the proprietary and FAANG-level generative AI sector now seems to be so cautious of potential criticism referring to the human synthesis capabilities of its tasks, that precise folks hardly ever seem in mission pages for main bulletins and releases. As an alternative, associated publicity literature more and more tends to point out ‘cute’ and in any other case ‘non-threatening’ topics in synthesized outcomes.

With the appearance of Hunyuan LoRAs, for the primary time, the group has a chance to push the boundaries of LDM-based human video synthesis in a extremely succesful (moderately than marginal) system, and to completely discover the topic that the majority pursuits the vast majority of us – folks.

Implications

Since a seek for ‘Hunyuan’ on the Civit group principally reveals superstar LoRAs and ‘hardcore’ LoRAs, the central implication of the appearance of Hunyuan LoRAs is that they are going to be used to create AI pornographic (or in any other case defamatory) movies of actual folks – celebs and unknowns alike.

For compliance functions, the hobbyists who create Hunyuan LoRAs and who experiment with them on various Discord servers are cautious to ban examples of actual folks from being posted. The fact is that even picture-based deepfakes are actually severely weaponized; and the prospect of including really sensible movies into the combination might lastly justify the heightened fears which have been recurrent within the media over the past seven years, and which have prompted new rules.

The Driving Pressure

As ever, porn stays the driving drive for expertise. No matter our opinion of such utilization, this relentless engine of impetus drives advances within the state-of-the-art that may in the end profit extra mainstream adoption.

On this case, it’s attainable that the value can be greater than typical, for the reason that open-sourcing of hyper-realistic video creation has apparent implications for prison, political and moral misuse.

One Reddit group (which I cannot title right here) devoted to AI era of NSFW video content material has an related, open Discord server the place customers are refining ComfyUI workflows for Hunyuan-based video porn era. Day by day, customers put up examples of NSFW clips – lots of which may fairly be termed ‘excessive’, or a minimum of straining the restrictions acknowledged in discussion board guidelines.

This group additionally maintains a considerable and well-developed GitHub repository that includes instruments that may obtain and course of pornographic movies, to supply coaching information for brand new fashions.

Since the most well-liked LoRA coach, Kohya-ss, now helps Hunyuan LoRA coaching, the obstacles to entry for unbounded generative video coaching are decreasing every day, together with the {hardware} necessities for Hunyuan coaching and video era.

The essential side of devoted coaching schemes for porn-based AI (moderately than identification-based fashions, similar to celebrities) is that a typical basis mannequin like Hunyuan shouldn’t be particularly skilled on NSFW output, and should subsequently both carry out poorly when requested to generate NSFW content material, or fail to disentangle realized ideas and associations in a performative or convincing method.

By creating fine-tuned NSFW basis fashions and LoRAs, it is going to be more and more attainable to mission skilled identities right into a devoted ‘porn’ video area; in any case, that is solely the video model of one thing that has already occurred for nonetheless pictures over the past two and a half years.

VFX

The large improve in temporal consistency that Hunyuan Video LoRAs supply is an apparent boon to the AI visible results business, which leans very closely on adapting open supply software program.

Although a Hunyuan Video LoRA strategy generates a whole body and surroundings, VFX firms have nearly actually begun to experiment with isolating the temporally-consistent human faces that may be obtained by this technique, to be able to superimpose or combine faces into real-world supply footage.

Just like the hobbyist group, VFX firms should look ahead to Hunyuan Video’s image-to-video and video-to-video performance, which is probably essentially the most helpful bridge between LoRA-driven, ID-based ‘deepfake’ content material; or else improvise, and use the interval to probe the outer capabilities of the framework and of potential variations, and even proprietary in-house forks of Hunyuan Video.

Although the license phrases for Hunyuan Video technically enable the depiction of actual people as long as permission is given, they prohibit its use within the EU, United Kingdom, and in South Korea. On the ‘stays in Vegas’ precept, this doesn’t essentially imply that Hunyuan Video is not going to be utilized in these areas; nevertheless, the prospect of exterior information audits, to implement a rising rules round generative AI, might make such illicit utilization dangerous.

One different probably ambiguous space of the license phrases states:

‘If, on the Tencent Hunyuan model launch date, the month-to-month lively customers of all services or products made accessible by or for Licensee is bigger than 100 million month-to-month lively customers within the previous calendar month, You could request a license from Tencent, which Tencent might grant to You in its sole discretion, and You aren’t approved to train any of the rights beneath this Settlement except or till Tencent in any other case expressly grants You such rights.’

This clause is clearly aimed on the multitude of firms which are more likely to ‘intermediary’ Hunyuan Video for a comparatively tech-illiterate physique of customers, and who can be required to chop Tencent into the motion, above a sure ceiling of customers.

Whether or not or not the broad phrasing might additionally cowl oblique utilization (i.e., by way of the supply of Hunyuan-enabled visible results output in standard motion pictures and TV) might have clarification.

Conclusion

Since deepfake video has existed for a very long time, it could be simple to underestimate the importance of Hunyuan Video LoRA as an strategy to identification synthesis, and deepfaking; and to imagine that the developments at the moment manifesting on the Civit group, and at associated Discords and subreddits, symbolize a mere incremental nudge in the direction of really controllable human video synthesis.

Extra doubtless is that the present efforts symbolize solely a fraction of Hunyuan Video’s potential to create utterly convincing full-body and full-environment deepfakes; as soon as the image-to-video element is launched (rumored to be occurring this month), a much more granular stage of generative energy will change into accessible to each the hobbyist {and professional} communities.

When Stability.ai launched Steady Diffusion in 2022, many observers couldn’t decide why the corporate would simply give away what was, on the time, such a worthwhile and highly effective generative system. With Hunyuan Video, the revenue motive is constructed instantly into the license – albeit that it could show tough for Tencent to find out when an organization triggers the profit-sharing scheme.

In any case, the outcome is similar because it was in 2022: devoted improvement communities have fashioned instantly and with intense fervor across the launch. A few of the roads that these efforts will take within the subsequent 12 months are absolutely set to immediate new headlines.

* As much as 136 by the point of publication.

First printed Tuesday, January 7, 2025