A newly formed open-source artificial intelligence developer alliance is ambitiously planning to launch a new AI model that may directly compete with Stable Diffusion and its controversial latest version.
Last week, the professional studio’s Generative AI platform Invoke, Comfy.org focusing on ComfyUI development, and the world’s largest stable diffusion repository CivitAI launched the Open Model Initiative (OMI). Now, the organization is leading a community-driven effort to develop open-source AI models for image, video, and audio generation.
Invoke AI CEO Kent Keirsey told Decrypt, “I hope OMI will launch a new next-generation model that the community can rally around and build support for in the next three to six months – it’s an ambitious goal for a new organization, but I believe we can achieve it.”
OMI aims to develop open-source models that are equal to or higher in quality than proprietary models such as Ideogram or MidJourney, but without the restrictions of proprietary license terms.
CivitAI claimed in a recent post, “Comfy tends to use a better-designed SD3.”
This move comes shortly after the release of Stability AI’s SD3, which has faced criticism for its restrictive licensing terms. While the stable diffusion series of AI models is very popular, the SD3 license is considered too limiting, leading to its ban by CivitAI.
To prevent this from happening again, the organization advocates for true open-source development and may adopt the MIT or Apache-2 license to ensure that truly open, unrestricted models meet minimum requirements.
Members stated in an open letter, “We believe open-source is the best way to ensure that AI benefits everyone. Through collaboration, we can offer high-quality, competitive models with open licenses, drive creativity in AI forward, and meet the needs of the community, free of charge.”
To address ethical concerns, the initiative is also committed to developing a base model with no pre-trained capabilities, such as “recognizing unauthorized artist names” and “generating unauthorized personal portraits.”
The initiative has received a large amount of support, with over 1,000 members joining its Discord server. Keirsey told Decrypt, “We’ve received over 100 requests for joining and support in the past 24 hours alone.”
Regarding funding, Keirsey stated that the initiative will not seek venture capital, relying instead on community support and the business model of founding members.
“OMI will not accept investment, as the purpose of the initiative is to establish open models, not to generate profit,” he said. “We’ve already seen what happens when profit becomes the motivating factor behind organizations based on open access to AI.”
Each member of OMI will maintain their own business structure, despite their involvement in the group.
Robin Ken of ComfyUI told Decrypt that their role in the mission of “democratizing AI” is at the tool level, unlike other initiatives more focused on developing models such as Stability, Mistral, or Meta. He also confirmed that Comfy is not just a side project, with financial backing.
“Comfy will always remain open-source,” Ken told Decrypt. “We have venture capital support and plan to make money from consulting/business support.”
Ken, Alex Goodwin, and the co-founder Comfy Anonymous left Stability AI to focus on the development and growth of Comfy.org with other developers.
The Open Model Initiative stated that it will initially focus on organizing volunteers, determining governance structure, and managing datasets with community assistance.
LAION is an AI company that compiles image and caption datasets scraped from the web for training several stable diffusions. Initially announced as a founding member of OMI, LAION was later removed.
Edited by Ryan Ozawa.
