Images and videos generated by artificial intelligence models are slowly approaching the level where, without very thorough analysis, it becomes virtually impossible to distinguish them from a real recording or photo. This is a growing problem for at least several reasons – not only related to disinformation, but also related to a fundamental change in thinking about visual art. So far identified solely with human creativity, today it is no longer as clear.
The recent official disclosure of DLSS 5 technology from Nvidia, through the use of additional light sources and correction of original textures, caused quite a stir among both creators and recipients. A moment earlier, the debut of the Seedance 2.0 model greatly shocked Hollywood and forced entities such as Disney and Paramount to take legal action. The newly debuted ChatGPT Images 2.0, which is currently the best model for generating photos (at least until the debut of Nano Banana 3 from Google), is also having a major impact on the creative industry… and more groundbreaking tools are already on the way.
On average, once every few weeks, big tech makes another breakthrough that takes the generation of images and videos to an even higher level. What does this really mean for the film and computer graphics industries? Is AI “taking their jobs”? Or do they look at it more as an opportunity? We asked several questions about technological progress to Platige Image – a Polish company specializing in this area and operating on this market for almost 30 years.
Łukasz Gołąbiowski, Business Insider Polska: Have you ever seen an AI-generated clip that made you think that “this is really good” or “this is at the level of what we do ourselves”?
Mateusz Tokarz, Head of CG at Platige Image: Yes, we've seen AI clips that make you say “this is really good”, but we don't yet regularly see public examples that overall measure up to the best CG/VFX productions. However, it should be remembered that presentation materials are usually prepared for specific presentations, i.e. if the model was mainly trained on racing films, the best content will be created on racing topics and these will also appear in promotional materials. However, it cannot cope with human action and we will not see it until we try to generate something like that ourselves.
Of course, this is not an argument against using AI – it is a tool like any other, but you need to remember that it has limitations and know how to use them. Another thing is that the market has started to accept and pay for materials that can often be described as “happy accidents”.
You keep up to date with changes in models such as Seedance, Veo and Kling. Progress in this area has been enormous in recent months, but it is still not quite there. Which elements, from your perspective, are still the most different compared to materials produced in a classical way?
The model manufacturers' announcements themselves show very well where their weaknesses still lie. Google in Veo 3.1 highlights “consistency, creativity and control”, reference images and 1080p/4K upscaling. Kling 3.0 emphasizes multi-shot storytelling, consistency and native audio. Seedance 2.0 talks about the increase in physical accuracy, controllability and 15-second, multi-shot audio-video clips, and at the same time directly admits that the model is “still far from perfect” and has visible artifacts.
The marketing of these products simultaneously describes their greatest advances and greatest gaps. A huge problem of models is still the lack of subtlety of movement (which results directly from how they technically work), its continuity and controllability, which also translates into a lack of continuity between shots.
See also: Vibe coding and the vibe of the apocalypse. We are preparing a digital Chernobyl ourselves
From our perspective, especially in the area of cinematics, the most impressive thing is not the content generation itself, but the impact of these tools on the entire production process. We see the greatest value in how AI improves and accelerates subsequent stages of work – from preparing storyboards and previs, through the creation and iteration of assets, to animation and post-production. This really changes the dynamics of production, allowing you to test ideas faster and better use the team's time.
Thanks to this, we are able to devote more energy to quality and refining the final effect, instead of time-consuming technical operations.
At the same time, it is worth emphasizing that, as in other areas, AI tools still require a selective approach and very conscious quality control, especially when we care about high visual value.
Which creative processes then “gained” and which “lost” the most from the inclusion of artificial intelligence in them?
The stages in which AI shortens the iteration but does not take responsibility for the final frame have gained the most. It helps the most where it can support repetitive and technical tasks – to speed up workflow and shorten the time of performing activities that are unnecessary in the process and exist only due to existing technological shortcomings.
We are talking about tasks such as denoise and upscale in rendering, rotoscoping, tracking, face replacement, cleanup, motion capture data cleaning, animation retargeting, background and matte painting generation, as well as some pipeline and organizational tasks. However, it causes harm where it tries to replace the artist – by duplicating the so-called AI slops and gives a false sense of control over the creative process. In addition, there is the issue of uncertainty around copyrights to materials generated by AI (although this may change – here I refer to the Academy Software Foundation project)
Well, in the creative industry the topic of “morality” of using genAI is particularly important – how do you approach it? Do you have any internal rules regarding the use of AI in your daily work that you would also recommend to others?
We have no problem with AI as a workshop accelerator; we have a problem with AI used without consent, without transparency of data origin and without a human being who takes responsibility for the effect. In our daily work with AI, we are guided by several important principles:
- We primarily use tools that make it possible to defend the origin of the data and the chain of rights;
- We do not use a specific person's voice, face, movement or style without informed consent and clear remuneration policies;
- We do not train anything on client assets, artists' work or performance materials without consent;
- We document AI participation when it is important for the client, partner or recipient; the framework for transparency is already set by Art. 50 AI Act.
- It is humans who create and decide – even using AI tools – we combine classic technology and AI where it can actually help speed up the process and not replace creative people.
A recently published report after the last GDC conference indicated that as many as 52 percent developers negatively perceive the impact of AI on their work. This is a significant increase compared to the results from last year, the trend seems obvious. How does your team view the progress of technology in this area and in what processes are you already using it?
This result doesn't surprise us, as any technology that affects the way we work naturally raises concerns – especially in the creative industry. Many of them result from uncertainty as to the role of humans in the process and the quality of the generated results.
From our perspective, AI is now primarily a tool to increase productivity and enable faster iteration. We are already using it, among others: in the areas:
- pre-production and lookdev
- automation of selected pipeline elements
- processing and editing animation data
- prototyping technical solutions.
The key thing, however, is that AI does not replace creative decisions, and in fact sometimes it can help make them more consciously. The greatest value remains the team's experience, artistic sensitivity and ability to make conscious choices.
Nvidia recently showed a technology that caused quite a stir due to how much it interferes with the graphics presented on the user's screen. Do you share emerging concerns about DLSS 5 distorting artistic vision?
This is a justified concern, but in our opinion it is mainly due to the fact that we are talking about technology that directly interferes with the final image. Any processing of this type carries the risk of artifacts or changes in the perception of detail.
Therefore, it is crucial to treat such tools as an optional technological layer and not the foundation of the artistic vision. Responsibility for the final appearance still rests with the creators and they must have full control and the ability to validate the final effect. Even if this effect is processed by AI “filters”, the creator should be able to consciously decide whether his work should use them or not.
At the stage of presenting this new image enhancement technology, your studio was not associated with it in any way, but it obviously applies to your business as well. Has Nvidia already approached you with any proposals for cooperation in this area?
It is worth distinguishing two things: marketing communication and the actual use of technology in production. Lists published on the occasion of the premiere of new solutions usually concern implementation directly in games. In the area of cinematics, the situation is different. Technologies of this type can be used as part of a pipeline, rather than as a function exposed to the end user.
In practice, this means that the lack of presence on the official lists does not necessarily mean a lack of interest or tests. It's just that in the case of film productions and cinematics, specific technologies behind the process are less often communicated – more emphasis is placed on the final effect.
Decisions about the possible integration of such solutions always result from a very pragmatic assessment – whether a given technology actually supports the quality and efficiency of production in a specific project.
How do you assess the usefulness of this type of tools for your productions and to what extent do you plan to use them once they become more widely available?
Technologies such as DLSS 5 have real potential to increase production efficiency, but at the current stage they must be used very selectively. This applies more broadly to the entire class of AI-based tools.
In practice, we still see typical quality problems – especially in areas related to image perception, where the so-called uncanny valley. Even minor artifacts or inconsistencies can be very noticeable to the recipient. The audience is already used to “classic” CG and even if it is not photorealistic, it does not spoil the reception. One can see an analogy of AI to the Polish voice-over in foreign films, which is impossible for people who did not grow up watching such films.
See also: CD Projekt RED is getting ready to release The Witcher 4. The competitor is the most important game in history
Therefore, we treat these types of solutions as supporting tools that require strict control and validation. Their use depends on the specific context – the scene, the type of content and the level of quality expected. These are not yet technologies that can be used completely unconditionally. Although they probably will be someday.
As we are slowly approaching the end, I wanted to ask one more thing about the gaming industry – will we see any new trailers of your creations* this year? If so, when can we expect them?
There are several very different projects in the pipeline at the moment – they will appear online in the coming months, in line with the developers' marketing plans.
And outside the gaming industry? Which of your current projects do you see as the most intriguing and the most challenging for the studio?
For us, each project is, above all, an incredible adventure – an opportunity to enter the world of a game, film, series or amusement park and meet new creators, being infected by their energy and vision of the world they create. The biggest challenge for a long time has been not the design, but the market, which has become highly unpredictable. It's definitely not permanent, but you have to survive this period.
* – Platige Image has recently responded, among others, to: for the first cinematic trailer for The Witcher 4 from CD Projekt RED
