AI in Optimystic

For over a year our studio has been developing an internal division “Optimystic Lab”. It focuses on research, development and implementation of neural network tools and custom pipelines for processing video content and individual images.

EXPERTISE

The neural network market is developing rapidly: what seemed impossible yesterday is now realizable, and tomorrow's opportunities are becoming available ahead of expectations. The Optimystic Lab team monitors the latest releases on a daily basis, tests them on our own data and promptly introduces the most promising solutions into the production pipeline.

The nearest plans are to form a chain of neural network tools, which will make it possible to speed up and cheapen the production of commercials. To date, we have managed to achieve the quality of AI images and videos comparable to traditional filming, which used to require weeks of preparation, multi-day shoots and significant budgets.

Now, the same result can be achieved faster and much cheaper. Below we showcase the results of our research and AI combat projects.

Tools

Tool
Image generation
Midjourney
Flux
Stable Diffusion
Google ImageGen
Video generation
Kling
Runway
Luma Dream
Hailuo Minimax
Google Veo3
All in one
Krea
Freepik

Each neural network has its own tasks and strengths, so it uses most of the actual tools available today. Some are strong in control, some in creativity, and some in image detailing. The best fit for the task based on our experience and research is used. We are always on the lookout for updates and as soon as something new comes out, it is immediately put into a testing period or used right away to solve problems.

As for Google Veo3, it and other neural networks from Google: Gemma (LLM) or ImageGen4 are used by us in our work.

Case studies. Creative development

ENVIRONMENTAL DESIGN

In the past, artists used to assemble backgrounds by photobash or hand-draw locations. Now we generate dozens of photorealistic environment variants in a matter of hours, from which the client chooses a direction.

CHARACTER DESIGN

Whereas it used to take concept artists, 3d modelers and visualizers several weeks to develop characters, today we can develop characters in a matter of days, going over many options with the client.

SEARCH FOR COLOR AND LIGHT SCHEMES

With the help of properly selected references and artificial intelligence, we can quickly decide on the overall bow and mood of the future video.

1 / 4

Case studies. Production

FULL-FEATURED VIDEOS

We can completely generate a 30-second commercial without filming, which saves a huge amount of money and time for the client. We managed to achieve consistency in the shapes, color and dynamics of the frames of the commercial, which allows us to get a complete product, difficult to distinguish from classic commercials.

SHOTS REPLACING EXPENSIVE SHOOTS

Scenes that in classic production required renting pavilions, robotic cameras or complex TableTop equipment, we create entirely with generative models.

KEY VISUALS

Based on photographic material of real people, we create stylized or realistic high-resolution images, ideal for any purpose, saving the client from expensive and lengthy photo shoots.

DIGITAL DOUBLERS

Based on the actor's photos, we train a model and get an “AI-copy” of him that can be shown in different costumes, eras and locations.

UPSCALE

If necessary, we upscale original footage or archival footage to 4K/8K without loss of detail using super-resolution models.

Legal details

When using neural networks for commercial purposes, we are sure to:

Check licenses

Clarify whether the neural network allows commercial use, and comply with the terms specified in its license

Fix the terms

In the contracts we must formulate the terms of transfer of rights taking into account the specifics of the neural network interface used in the work (for example, granting a non-exclusive license for the generation results)

Avoid violations

Do not generate images similar to protected images, or obtain confirmation from the client that they have the rights to such an image and the rights to use it

Respect the authors

Avoid directly copying the styles of famous artists or brands

Create a unique

We use custom prompts to train the neural network and generate unique (to the extent possible, given the specific use of AI) images and videos

Focus on the law

We are based on the Russian legislation (in particular, Article 1257 of the Civil Code of the Russian Federation, according to which the author is a natural person whose creative labor created the work)

Publishing AI videos on social media

Key requirements to AI content
Instagram / Facebook
From May 2024, Meta will automatically label photos, videos and audio with the “Made with AI” tag. Publications and advertisements with generative content must disclose the fact of synthetic origin, especially if the material “significantly alters reality” (deepfake) or depicts non-existent people
VKontakte
There is no direct ban on AI videos, but the general rules on the protection of the citizen's image and the prohibition on the dissemination of inaccurate material apply

Limitations

Inconsistent results

With repeated iterations of prompts, neural networks can generate fundamentally different images or videos, even with minor adjustments. It is still difficult to control changes locally (correcting only a detail)

Unpredictable video generation

Each new attempt creates virtually a new clip; it is virtually impossible to achieve spot edits within a frame

Production paradigm shift

Instead of traditional deep elaboration of the selected version, we create several variants (options) of the already final elements at once and choose from them, as it is difficult to detail one option without noticeable deviations

Insufficient camera control

Generative video does not allow you to control camera movement as precisely as you can on a movie set or in a 3d scene

Difficulty with emotions

To date, it is very difficult to give an AI actor the right emotional emphasis, and if a commercial involves acting, AI involvement is problematic

Unavailability of small texts

When generating packages, labels and other materials with fine typography, neural networks are not yet able to create realistic readable inscriptions - the text either turns into an abstract set of symbols or loses readability when enlarged

The final chord

We live in an era of profound shifts in the content production industry. Methods that have remained unchanged for decades are now transforming before our eyes. Artificial intelligence is changing the rules of the game: reducing costs, shifting deadlines, and opening up previously impossible visual solutions.

We are not afraid of the AI era - on the contrary, we embrace it with enthusiasm. For us, AI is not a replacement, but an enhancement: it expands creative horizons, strengthens the team's technical expertise, and allows us to focus on what matters most - the idea.

All images and videos used on this page are generated by Optimystic Lab