A bit of context would be nice. I have literally no clue what I am looking at.
replyThese are not meaningless.
replyIf you do regular image gen, you can see that these images demonstrate memorized styles, broken anatomy, bad text layout (or hallucinated text), etc. Even the old Dall-E watermark is a throwback.
Generation times are orthogonal.
I don’t do image gen regularly… so I’d love more context. I can see some less good features in the images but I’m not fluent with what today’s generators would do. I agree with GP that it would be quite useful to know about the prompts & iteration effort. If generation times changed, they could be relevant & helpful.
replyI don't do extremely regular image gen, but I've been running ComfyUI on my 3090 for, jeeze, two years?
replyHere's my homelab blog: https://www.gavmor.com/regular-labs/
Maybe I'm just not that observant, but all these pictures are just... "LGTM."
When I'm generating, I have really specific adherence/performance goals.
This quickly escalated from nostalgia to ComfyUI image optimization
replySpring 2022 was not an adherence/performance era.
replyI think we can still appreciate these snapshots in time.
In the future, this will be an aesthetic. Some already consider early AI jankiness a style.