A newly formed alliance of open-source AI developers has ambitious plans to launch a new AI model, one that could compete directly with Stable Diffusion and its controversial latest release.
The Open Model Initiative (OMI) was launched last week by Invoke, a Generative AI platform for Professional Studios, Comfy.org, which focuses on the development of ComfyUI, and CivitAI, the world’s largest Stable Diffusion repository. Now, the group is spearheading a community-driven effort aimed at developing open-source AI models for image, video, and audio generation.
“I hope to have the OMI launch a new, next-gen model that the community can rally behind and build support for in the next three to six months—an ambitious goal for a new organization, but one that I am confident we can achieve,” Kent Keirsey, CEO of Invoke AI, told Decrypt.
The OMI aims to develop open-source models of equal or greater quality than proprietary models (like Ideogram or MidJourney) but free of restrictive licensing terms.
“Comfy is inclined to leverage a better-designed SD3.” CivitAI declared in a recent post.
This initiative comes shortly after the launch of SD3 by Stability AI, which was criticized for its restrictive licensing terms. While the Stable Diffusion family of AI models has been massively popular, the SD3 license was deemed so restrictive that it led to its ban from CivitAI.
To prevent this from happening again, the organization advocates for truly open-source development and will likely adopt the MIT or Apache-2 license, which ensures truly open, non-restrictive models subject to minimum conditions.
“We believe open source is the best way forward to ensure that AI benefits everyone,” the members said in an open letter. “By teaming up, we can deliver high-quality, competitive models with open licenses that push AI creativity forward, are free to use, and meet the needs of the community.”
Addressing ethical concerns, the initiative also committed to developing a base model without pre-trained capabilities such as “recognition of unconsented artist names” and “generating the likeness of unconsented individuals.”
The initiative has garnered significant support, with over 1,000 members joining its Discord server. “We’ve fielded more than 100 requests to join and support in just the last 24 hours,” Keirsey told Decrypt.
Regarding funding, Keirsey told Decrypt that the initiative would not pursue venture capital, but rather rely on community support and the business models of its founding members.
“The OMI will not be taking investments, as the initiative’s aim is building open models, not generating a profit,” he said, “We have seen what happens when profit becomes the motivating factor behind organizations founded on the principle of open access to AI.”
Each member of the OMI will maintain its own business structure despite its involvement with the group.
Robin Ken of ComfyUI told Decrypt that their role in the mission of “democratizing AI” is at the tooling layer, unlike other initiatives like Stability, Mistral or Meta, which are more focused on developing models. He also confirmed that Comfy is more than a side project and has financial backing behind it.
“Comfy will remain open source forever,” Ken told Decrypt. “We are VC-backed and plan to make money from consulting/enterprise support.”
Ken, Alex Goodwin, and pseudonymous cofounder Comfy Anonymous left Stability AI to focus on the development and growth of Comfy.org, along with other developers.
The Open Model Initiative said it will initially focus on organizing volunteers, deciding on a governance structure, and curating datasets with community assistance.
LAION—an AI firm that compiled datasets of images and captions scraped from the web that were used to train several Stable Diffusion—was initially announced as a founding member of the OMI but was later removed.
Edited by Ryan Ozawa.
Generally Intelligent Newsletter
A weekly AI journey narrated by Gen, a generative AI model.
Source: https://decrypt.co/237350/ai-open-model-initiative-new-compete-with-sd3
