T-DIG demo: ask for Halloween costumes, get food puns and ML researcher outfits instead of 8 witches
simi_97k · x · 2026-10-28
The T-DIG team demos the pain point with Halloween: asking an image model for "a person in a halloween costume" returns the same witch in 4-8 of 8 i.i.d. samples, and prior unconditional diversity methods just change faces, not costumes.
T-DIG lets you specify the variation axis: "a food pun" yields pumpkin, pizza, burger, and ice-cream costumes, while "low-effort costumes for ML researchers" produces byte me, data miner, neural net, and bug squasher. The tool works with any model and any request.
More from Fun
- Dev read vLLM codebase line-by-line with a debugger for hours — wouldn't do it again with AI agents — gordic_aleksa · 2026-10-28
- Founder warns Claude's 'moral policing' turns privacy policies into legal liabilities — StewartalsopIII · 2026-10-28
- One-line Tekken-style prompt in HeyGen Video yields ad-grade 2K fight footage — aziz4ai · 2026-10-28
- Explaining 2026 to my 2022 self: developers no longer read or write code — JnBrymn · 2026-10-28
- Teenage Engineering stops making synths — App Studio lets you build your own — c_valenzuelab · 2026-10-28
- Engineer compares working with AI agents to being married — yawnxyz · 2026-10-28