After watching Lu Yang: why AI circles talk models, not aesthetics
Professional-Cap-377 · reddit · 2026-08-20
A painter and art-history background author reflects on the gap between what digital tools can do and what most AI image/video work looks like. The key is art direction and having something to say—visual judgment comes from consuming art, film, photography, and animation, not from the tool itself.
They note that aesthetics, art direction, and visual language get far less discussion in AI communities than models, nodes, speed, resolution, and technical efficiency. ComfyUI could become a controlled artistic process—involving composition, 3D structure, lighting, references, masks, regional edits, repeated passes, and deliberate decisions about what the model may alter—rather than a content-generation machine. The real question is whether the masses using these tools can develop aesthetically at the same pace the tools improve.
More from Multimodal
- Hands-On: NoSpoon H3 Agent Turns Great Gatsby Screenplays Into Music Videos in 15 Minutes — Kyrannio · 2026-08-20
- Seedance 2.5 launches on Pollo AI with 30-second video support — HeyAmit_ · 2026-08-20
- Open-source browser extension swaps any web image in place via local ComfyUI workflow — CreepyInpu · 2026-08-20
- Can H3 use a reference image to upscale low-res faces in video? — ignoramati · 2026-08-20
- Can MMH3 generate 40+ second audio drama clips? — wh33t · 2026-08-20
- AI video creators self-roast: one thin prompt line, everything else is correction — CurieuxExplorer · 2026-08-20