Hacker News Reader

at this point perfect text would be a gamechanger if it can be solved

midjourney 6 can be completely photorealistic and include valid text, but also sometimes adds bad text. it's not much, but having to use an image editor for that is still annoying. for creating marketing material, getting perfect text every time and never getting bad text would be amazing

View on HN

I wonder if we could get it to generate a layered output, to make it easy to change just the text layer. It already creates the textual part in a separate pass, right?

Current open source tools include pretty decent off the shelf segment anything based detectors. It leaves a lot to be desired, but you do layer-like operations automatically detecting certain concept and applying changes to them or, less commonly exporting the cropped areas. But not the content "beneath" the layers as they don't exist.

Which tools would you recommend for this kind of thing?

comfyui + https://github.com/ltdrdata/ComfyUI-Impact-Pack

I would bet that Adobe is definitely salivating at that. Might not be for a long time but it seems like a no brainer once the technology can handle it. Just the last few years have been fast and I interacted with the JS landscape for a few years. It moves faster than Sonic and this tech iterates quick.

Reply on news.ycombinator.com