In 2024, AI was an equal partner in JetStyle’s everyday work tasks in all departments, from design to development and HR. We’re sharing our insights about work with neural networks on our website: https://jet.style/articles-ai
Alex Perminov, JetStyle’s PR Designer, is one of the ambassadors for experimenting with AI. We asked him to give an overview of what happened in the industry in 2024 and what he expects from 2025:
1. General trends
After significant advancements in generative text-to-image neural networks, we expected breakthroughs in video and 3D generation. However, the biggest progress this year came from text-to-text AI. What once felt like magic in ChatGPT now seems mundane compared to what GPT-4 or Claude do. DALLE-3 is catching up as well: it emerged as a strong competitor to Midjourney and Stable Diffusion, thanks to its integration with GPT's language model. DALLE-3 uses it for better understanding of user prompts and image generation accuracy.
2. AI Video
By the end of 2024, OpenAI released SORA, their long-awaited video generation model. However, a year after its initial announcement, the excitement about its features has significantly faded. Throughout the year, numerous competitors developed impressive AI-based video generation tools, with Chinese developers like Kling and Hailuo leading the pack.
As a user, I’m happy about this level of competition: it drives innovation and makes AI video tools increasingly viable for production.
3. New services
This year taught me not to get too attached to cool free tools — just like you wouldn’t get attached to a character in Game of Thrones. Many services that started as completely free moved to subscription-based models by midyear. Now they offer only a few free generations weekly; unfortunately, in some cases the quality of those has declined as well.
I used to hate ads on TV and on the internet, but now I miss the days when free tools were funded by ads rather than subscriptions. It was really a win-win situation for companies and users. What do you think?
4. Censorship & new regulations
On one side, open-source models generate things you can’t unsee; on the other, major players like OpenAI and Google suppress creativity with strict copyright and NSFW content filters. Rules are good, but it seems they bring more harm now. Google’s AI overreacts to users’ prompts, DALLE-3 struggles with generating art in specific styles due to copyright concerns. SORA is still in the process of fine-tuning their censorship requirements, and it remains an inconvenient tool.
We are so far from the state of balance in this matter, aren't we?
5. Switching to interfaces
I can’t remember the last time anyone complained about the horrors of the interface of Stable Diffusion. Most SD enthusiasts have moved to user-friendly alternatives like Forge, while PikaAI and Luma transitioned from Discord to web interfaces. Even Midjourney finally launched a proper website for web-based generations. It's definitely a positive change.
What’s next?
My predictions about AI never come true, so I’ll skip them this time. Instead, here’s what I’d love to see in 2025: a more consolidated market. When someone asks me for an AI tool recommendation, I have to look through a chaotic list of 100+ neural networks, many of questionable quality. It’d be fantastic if the weaker players phased out and left us with 3-5 reliable, versatile services.
Personally, I want to dive deeper into hardcore tools like Stable Diffusion and ComfyUI. I’m also looking forward to updates from Midjourney and hoping they don’t increase their subscription fees too much.
At JetStyle, we harness the power of AI to save resources and unlock new levels of creativity. Let’s create something extraordinary together: Google Calendar
