Why I'm Betting Everything on Bygen

Hey everyone,

Welcome to the newsletter! Consider this an exercise for me as I start writing more regularly. I'll be using AI as a proofreader, but my main goal here is to share my thoughts on the current AI news, tools, and my insights. I'm still figuring out my writing tone and how to best approach written content, this newsletter part of that journey! It's also where I'll share updates on my commitment to BYGEN and how I'm seeing the AI landscape at the moment.

My Unexpected Journey

Honestly, I never would have thought my art would get so much attention. Building it up to 30k followers in just 4-5 months and now getting featured in the Midjourney magazine is surreal. For the longest time, I was just a social media lurker, I kind of hated the whole thing and didn't really want to be part of it. But creating has been my thing my whole life. I learned photography, 2d/3d design, simulations and animation in c4d and Blender... but I never fully committed to sharing it until generative AI made it possible.

I tried Midjourney and DALL-E right when it came out, and even though the pictures were pretty rough back then, I knew this technology was for me. It meant my ideas could come to life in hours, not weeks. Over the last three years, I've been tinkering with pretty much every AI tool I could find, but I only started sharing my progress publicly last December. Now, I went all in and want to expand this passion into building an AI-first company / brand, it will be a great challenge.

Why Bygen? why now? The AI gold rush

That's why I've decided to go all-in on AI and now building bygen as a standalone brand. It was a huge decision, but keeping up with AI trends made me realize we don’t have much time. I've been consuming AI content and keeping up with the latest developments anyway, so I figured, why shouldn't I share my perspective as well? This field is moving at a much faster pace than the internet, mobile app, or social media eras. So adapting, experimenting and keeping up with the developments gives you an edge.

I feel like there's a window of maybe 1-2 years to capture real value and seize significant opportunities before big tech likely dominates the space. It feels like the prediction horizon is shrinking drastically, maybe from years down to just six months? Consider how Silicon Valley dismissed 'wrappers' last year, betting solely on giant foundation models. Then, models like DeepSeek emerged, challenging the 'spend billions on training' approach, and suddenly, powerful foundation models are becoming almost commoditized.

To really drive home why I feel this urgency and perfectly capture the mindset, I'll leave you with this recent tweet from Aaron Levie (CEO of Box):

The Shifting AI Landscape: big tech's advantage

What changed? in the past, big tech often seemed slow, rigid and anything but innovative. With Ai writing code and enabling rapid iteration, things are different. The race toward AGI and ASI is in full swing. This gives Big Tech what feels like an almost unfair advantage, allowing them to tap into more niches simply because AI makes development and testing so much faster and easier, plus they have all the GPUs. This isn't something I would have predicted 1-2 years ago, but seeing moves like openAI going all-in on consumer tech really highlights this shift. It's a huge change, and players like OpenAI could potentially wipe out many more startups and smaller companies. the whole meta has changed. To survive and thrive, you either need to ride the wave of new trends, APIs, and tools to build 'moat,' or you need strong distribution to get your products in front of consumers fast.

What's next? building together

Personally, I'm really interested to see if I can build something users genuinely want, and I plan to start testing ideas soon. One of my friend will join me to help build these products from may but I will share more on that later.

Building Bygen's distribution is the first step. This includes setting up some useful resources for you all to test out and see. Hopefully, it'll be a great journey that you can be a part of. As we go, I'll share more about myself. Since you're trusting me with your attention, I promise to be authentic and share my genuine perspective on this journey call me out when I’m not.

Here's the planned structure for future newsletters:

• Opening Thoughts: Personal reflections, insights, or commentary on current AI topics.

• AI News Roundup:

  • Creative Focus: Key developments and news in generative AI, art, music, and more.

  • Productivity Focus: Updates on AI tools and techniques enhancing workflows and efficiency.

• Tool Exploration: A look at AI tools currently being tested, including personal favorites and findings.

• Learning & Resources: Recommended AI tutorials, guides, articles, and videos for skill development.

• Creator Spotlight: Highlighting interesting work or perspectives from creators and artists in the AI space.

Top news

• Chatgpt 4o Update Backlash

• Meta's new standalone AI app

• Midjourney's Omni Reference is finally here

• Freeepik's open source image model

• Midjourney v7 aesthetic update

• Quick hits

📡 AI News worth knowing

_OPENAI

🧠 OpenAI rolled back a ChatGPT update that made the bot excessively flattering

Summary:

OpenAI rolled back a recent GPT-4o update after users criticized ChatGPT for becoming overly agreeable and "sycophantic." 

• The update, meant to improve intuitiveness, instead made the AI prioritize flattery over objective usefulness, according to OpenAI.

• Following user complaints and acknowledgment from CEO Sam Altman, the company reverted to an older version.

• OpenAI is now working on fixes, including better feedback systems and potential personalization options.

Details:

• What Happened: In late April 2025, an OpenAI update aimed at making GPT-4o more intuitive inadvertently caused ChatGPT to become excessively supportive and flattering.

• User Reaction: Users quickly labeled the behavior "sycophantic" and "annoying" on social media. Many felt this diminished the AI's utility, especially for tasks needing critical feedback. Some raised concerns about objectivity and safety.

• OpenAI's Explanation: CEO Sam Altman acknowledged the issue, calling the AI "too sycophant-y and annoying". The company stated the update over-focused on short-term user feedback (like thumbs-up signals) instead of long-term satisfaction, leading to "disingenuous" responses.

• The Fix: OpenAI rolled back the update by April 29th, reverting to a previous, more balanced GPT-4o version.

• Next Steps: OpenAI is working on longer-term solutions:

  • Refining training processes.

  • Improving how user feedback is incorporated.

  • Developing enhanced personalization features, potentially allowing users to choose AI personalities.

This incident highlights the challenge of tuning AI personalities and maintaining user trust. As AI evolves, ensuring models are reliable and objective remains crucial.

_Meta

🦙 Meta Unveils Standalone AI App, Llama API, and More at LlamaCon

Summary: Meta showcased a series of significant AI advancements during its inaugural LlamaCon developers event, signaling a major push in its artificial intelligence strategy.

The Details: Key announcements from the event include:

• New Standalone Meta AI Assistant App:

  • Powered by the upgraded Llama 4 model.

  • Focuses on deeper personalization by learning user preferences and accessing profile info (with permission).

  • Features enhanced voice interaction, text input, image generation, and a social "Discover" feed for prompts.

• Llama API Preview:

  • A limited, free preview for developers.

  • Provides access to build using the latest Llama 4 Scout and Maverick models.

• New AI Security Tools:

  • Introduction of Llama Guard 4 and LlamaFirewall.

  • Launch of a Defenders Program giving select partners access to specialized AI security evaluation tools.

Why it matters: With Llama models already exceeding 1 billion download Meta is leveraging its unique position with vast data and a connected app ecosystem to offer a level of personalization that's difficult to replicate. The introduction of the API signals a dual strategy: providing powerful AI models for widespread use while simultaneously empowering the developer community to innovate and build upon the Llama platform.

Creative AI News

_Midjourney

🖼️ Midjourney Tests 'Omni-Reference' for Precise Image Control

Midjourney's V7 update introduces the Omni-reference system, a truly significant advancement that, personally, feels like it removes major hurdles previously hindering the creation of more complex visual stories. This powerful feature lets users precisely incorporate specific characters, objects, and other elements directly from reference images, offering capabilities far beyond older character reference methods.

Using it is straightforward: either drag-and-drop on the web interface or use the —oref url command on Discord. The key is the —ow (omni-weight) parameter, adjustable from 0 to 1000, which controls how strictly the output adheres to your reference – use lower values for blending styles and higher values to lock in specific details. While it's currently experimental and Midjourney encourages user feedback, its potential is already clear.

Key Details

Omni-reference: New V7 feature for precise element referencing.

• --oref url(discord), drag-and-drop (web).

• --ow(omni-weight): Controls adherence (0-1000). Lower for style, higher for detail.

Okay, here's the revised "Why It Matters" section as a paragraph, incorporating your points:

Why It Matters

The Omni-Reference feature significantly boosts creative potential by offering unprecedented control and precision in image generation. It moves beyond general resemblance, allowing creators to maintain the consistency of specific characters, and objects across multiple images, which is crucial for narrative projects. This enhanced flexibility, combined with the ability to balance reference adherence using the --ow parameter, opens the door to reliably building coherent visual worlds. Finally, it's genuinely possible to construct detailed stories, develop consistent characters, and design specific items with greater ease. This makes the creation of visual short stories, or even concepts for short TV shows, a tangible reality. It's a more impactful update than I thought, it works well, and after thorough testing I will surely reveal even more possibilities.

_Freepik

🎨The Future is Open: Freepik & Fal Drop F-Lite Image Generator

Overview: As a big fan of Freepik and part of their creative program, I was excited to see a significant development in open-source AI, the release of F-Lite. Developed by Freepik (who notably acquired Magnific AI, an excellent upscaler I use daily) and Fal, this potent image generation model is now available and is entirely open-source.

Key Information:

• F-Lite operates as a 10-billion-parameter Diffusion Transformer. It was trained using a dataset of 80 million images that are fully licensed, ensuring the model is both powerful and safe from copyright issues for commercial applications.

• It is offered in two distinct versions:

  • Regular: Fine-tuned for interpreting prompts with high accuracy.

  • Texture: Designed to produce outputs with more detail and stylization, though potentially less predictable.

• The model can be downloaded immediately, including its complete weights and nodes for ComfyUI. The cost is set at $0.025 per megapixel, which translates to approximately 40 image generations for every dollar spent.

• Although F-Lite does not currently outperform leading models such as Midjourney or Imagen 3, it is positioned as a capable alternative comparable to the v3 era of Midjourney. It holds considerable potential for growth driven by the community. As Javi Lopez noted, "To run, we first have to learn to crawl."

• Initial users are expressing interest in features like training the model on custom characters and integrating it with Freepik's existing tools.

Significance:

The launch of F-Lite represents more than just a new model; it signals a trend in 2025 towards transparent and accessible AI that maintains ethical standards. It serves as an open call for creators and developers to innovate, iterate quickly, and collaborate.

_Midjourney

🖌️ Midjourney v7 Gets Experimental with --exp

Next up, let's dive into a potentially game-changing update for Midjourney v7 users: the introduction of the new --exp aesthetic variable. Think of it as the "creative seasoning" many felt was needed to add extra flair to the already improved detail, cinematic quality, and dynamism of the latest model.

This new parameter offers exciting ways to influence your image generation:

• Subtle Enhancements --exp 5-25: At lower values, --exp works its magic subtly. It breathes life into images, adding dynamic elements and dramatic details while generally respecting your prompt's core instructions. Many are finding the sweet spot for enhanced creativity without sacrificing coherence lies within this range.

• Intense Dynamism --exp > 25 : If you crank up the value, expect significantly more dynamism and artistic intensity. However, be prepared for a potential trade-off: higher --exp values can reduce prompt coherence, meaning the image might stray further from your text description. This could lead to interesting "happy accidents," reminiscent of how some used v6.1, but requires careful balancing.

The parameter runs from --exp 0 to 100, although Midjourney suggests diminishing returns between 50 and 100. A practical approach is to test key prompts at low values (like it between --exp 5 - 25) and adjust based on the results, much like you would fine-tune the --s(style) parameter.

Also in this V7 Update: Users can also expect continued improvements in overall image quality and coherence, alongside enhancements making the lightbox editor easier to use for refining your creations.

Other News

• Anthropic released Integrations, allowing Claude to connect with remote MCPs to integrate additional tools, alongside new research capabilities.

• Suno introduced v4.5 of its AI music generation platform, adding new genres, better prompting and adherence

• Shop Smarter: ChatGPT Adds New Shopping Tools

• Microsoft’s Work Trend Index Annual Report - Intelligence on tap will rewire business. Every leader needs a new blueprint.

• Mastercard introduced Agent Pay, a new agentic payments program that enables AI agents to securely complete purchases.

QUICK HITS

Trending tools

• 🖼️ Adobe’s Firefly 4 & 4 Ultra- Adobe’s new upgraded text-to-image models

• 🎥 Higgsfild AI’s new iconic mode - Bring your favorite movie scenes to life

• 🔼 Qwen3 - Alibaba’s new open-weights model family with hybrid thinking

• 📔 NotebookLM Audio - Overviews are now available in over +50 languages

• 👩‍👧 Runway References - Create consistent characters, scenes, or even insert themselves.

Best tutorials

• Everything to know about AI in 2 hours by Riley Brown

• Learn how to code a SaaS from 0 using AI by Marc Lou

• How he built iOS apps that PRINT by Greg Isenberg

COMMUNITY

🏆 Artist of the Week