I am trying my best to get the beta out as early as possible. I just need to get hold of some GPU to train the larger model to make it more generalised and usable for a public release. I am either going to apply for some grants or do some consultancy project or maybe even raise funding.
As mentioned in another comment I am using LLM, fine tuned text to image and a model tagging components. I haven't been able to get GPT4 API access as of yet but have applied and plan on shifting the LLM to GPT4 once I receive the access.
It is generative. I have used a combination of LLM, fined tuned text to image models and a model that tags the components from design systems. Post this I am writing code to convert this to figma components using figma API. Also using some heuristic algorithms to correct for layout artefacts.
I don't know about 100x faster app deployments but I am sure it can help create faster designs for freelance projects. Trying to understand the needs of complex designs in enterprises and how it can be incorporated to so solve more complex design problems.
Right now design system has to be manually defined as in a particular file and components in the design system have to named a certain way but I am trying to automate the process so that this manual effort doesn't need to be done. But maybe I can let it be like this also as it is only a one time effort and I am planning on providing several pre-named starter design systems.
Automating that would be helpful but I'm not sure how. Similar Text2SQL models require full description of each fields in the database. That is somewhat automated but in real world systems, it still requires manual explanation since they have the context on it.
Yeah I am also not exactly sure how I would go automate that process. Maybe I can use the figma code of the design system component and then run it through some classifier to classify it. I will work on it as soon as I release the beta.
Hey everyone for the past couple of months I have been working on a design tool that generates designs from prompts based on a design system.
As a UI engineer of 5 years I have been obsessing over the fact that if AI is generating designs then they should be based on a design system, much like co-pilot suggests code based on the pre existing codebase.
If you're interested in trying it out, you can sign up for early access on http://magify.design. Would be really glad to hear your feedback and thoughts on this tool, and I hope it can be used to take out the tedious parts of design.
