Cards against Cogniwerk: The Battle for Creative Freedom

As a startup, it’s vital for us to tackle our weaknesses, learning to fall and rise repeatedly. To accelerate this process, we’ve decided to confront our fears by asking experts in the creative industry to critique our service and offer suggestions for improvement. Every piece of feedback helps us grow, so thank you for your participation. Enjoy reading, and if you'd like to contribute to this series, please reach out to us at cardsagainst@cogniwerk.ai

Big Tech vs. Open Source: The Battle for Creative Freedom

by Peter, founder of Cogniwerk

In the realm of rapid AI development, a concerning trend is emerging: knowledge monopolization. Companies like OpenAI, in collaboration with Microsoft, are dominating the AI landscape, pushing the narrative that only infinitely large, expensive, and centrally hosted AI models are worthwhile. Their message: "Hand over the reins, we've got this. You just be the obedient users."

Google offers alternatives, but their offerings are just as closed off. The rules governing these models' training and application remain hidden from users, as do the identities and justifications of those who establish these rules.

Similar developments are unfolding in the creative industry, where Adobe and Midjourney reign supreme in image generation and editing. Despite their contrasting backgrounds (Adobe, a corporation, and Midjourney, a startup), they both claim monopolies.

Midjourney, founded in 2021 by David Holz, has established an seemingly unassailable position in generative image generation. Their tool entices creatives with its simplicity: a short prompt can yield surprisingly impressive results. The catch? The resulting images share a monotonous aesthetic, deeply embedded within the software.

When a tool like Midjourney exerts such profound influence on aesthetic output, questions arise: Who is responsible for this aesthetic? Holz himself? A training and reinforcement rules committee that essentially dictates the world's aesthetic direction? Who decides which terms are blocked and yield no results?

Users may scoff at this censorship on social media, but they fail to question their creative autonomy. Are they truly making independent choices, or have they fallen into the digital age's "convenience trap"? We surrender our autonomy to (mostly American) corporations, lured by the promise of effortless solutions. A dangerous trend, indeed.

This surrender threatens individual creatives, who become increasingly reliant on Midjourney's algorithm. It also jeopardizes the world of ideas and visions. The diversity of aesthetic perceptions across cultures is a treasure, akin to nature's biodiversity, and must be safeguarded.

These concerns fueled the creation of CogniWerk. There must be an alternative. And that alternative is open-source. We aim to empower users with control over their prompts and the aesthetic direction of their outputs.

We recognize the existence of hundreds of thousands of AI models, each trained with unique aesthetic and formal qualities, based on open-source Stable Diffusion. By selecting and utilizing these models, we can harness this diverse range of design perspectives and counter the monotonous uniformity of Midjourney and Adobe Firely's single-model outputs.

CogniWerk strives to operate on equal footing with its users, granting them control and enabling the creative freedom necessary for truly unique work. (Side note: Independent decisions are often less convenient than pre-made solutions. If you want to avoid fast food, you have to cook.)

CogniWerk envisions users bringing their own LoRa-formatted models and seamlessly integrating them into a unified user interface, designed to strike the perfect balance between ease of use and control for creatives.

We've made significant progress, but our interface still falls short of that sweet spot, remaining overly complex in many areas.

I love exploring Civitai.com, discovering the wildest LoRAs – fine-tuned models trained with specific moods, content, and characteristics, often created by amateurs and uploaded to the platform, but also by professionals from agencies and content generation experts, offering little wonders for our use.

How wonderful it would be if CogniWerk's "Bring your own model" promise became a reality: Copy a Civitai URL, paste it into CogniWerk, and voila! You can harness the power of your chosen LoRa masterpiece.

But alas, between this often exhilarating experience and CogniWerk.ai lies an interface that fails to clearly indicate where to input the LoRa or provide essential details (like the prompt's trigger word). Consequently, it often takes multiple attempts to successfully transfer a model from Civitai.ai to CogniWerk and utilize it effectively. Too time-consuming.

Therefore, I plead: CogniWerk, urgently enhance the handling of imported AI models! In the face of monopolistic threats, diversity and aesthetic freedom are paramount!

David versus Goliath – yes, but David needs to upgrade his arsenal!