Hacker News Reader

I'll take prompt adherence over quality any day. The machinery otherwise isn't worth it i.e the controlnets, openpose, depthmaps just to force a particular look or to achieve depth. Th solution becomes bespoke for each generation.

Had a test of it and my option is it's an improvement when it comes to following prompts and I do find the images more visually appealing.

View on HN

Can we use its output as input to SDXL? Presumably it would just fill in the details, and not create whole new images.

I was thinking that exactly. You could use the same trick as the hires-fix for an adherence-fix.

Yeah chain it in comfy to a turbo model for detail

A turbo model isn't the first thing I'd think of when it comes to finalizing a picture. Have you found one that produces high-quality output?

For detail, it'd probably be better to use a full model with a small number of steps (something like KSampler Advanced node with 40 total steps, but starting at step 32-ish.) Might even try using the SDXL refiner model for that.

Turbo models are decent at low-iteration-decent-results, but not so much at adding fine details to an mostly-done image.

Reply on news.ycombinator.com