Skip to main content


This generative model allows you to sketch out a scene with a few words, it then leverages an LLM to flesh out the details, with the ultimate goal of feeding those details to a downstream visual image generation model.
It is almost, but not quite, entirely the inverse of image captioning models.
This offers the closest experience to an image generation tool that's usable by people with visual impairments.

huggingface.co/spaces/lllyasvi…

#a11y #genai #llm

Tamas G reshared this.

in reply to Musharraf

Can this itself give you a downloadable image? I'm trying to work that out, and I found the 'Render image' button but I don't know that it does anything noticeable.
in reply to Andre Louis

@FreakyFwoof
Yes, it can generate actual images. Just click that button and wait for a little while.
It seems to load a separate model on demand when you click the button, which takes time.
in reply to Andre Louis

@FreakyFwoof
The image is added to the chat history with the role of button, and a standard label.
in reply to Musharraf

I got it, but couldn't save it even if I right clicked which was sad.
in reply to Andre Louis

@FreakyFwoof
I was able to save the generated image by focusing the entry containing the image, moving the mouse cursor to it, performing a right-click, and clicking "save image as".
in reply to Musharraf

I'll try again. I saved the output of the conversation to a file so maybe I can throw that back in, hit render, and try again.
in reply to Musharraf

Huge thanks for making this!

I'm wondering, how is this different compared to generating an image via Stable Diffusion for example? Are the images more predictable, i.e. closer to what your prompt is?

in reply to Erion

@erion
Sadly, I didn't make this. This is way beyond my resources.
But thanks for the thought 😁
in reply to Erion

@erion
The main difference is that this provides a detailed textual description. It doesn't generate the actual image.
You give it a simple prompt of what you want, and it draws the scene textually in full details.