Browsable library tests MiniMax H3 Image across 288 style prompts with one base photo
Numerous_Access_8008 · reddit · 2026-11-12
A Reddit user built a browsable style library for H3 Image: the same base photo (man in a café) rendered with 288 different style prompts—anime, painting, photography, comics, 3D, posters—same seed for clean A/B comparison. It supports category filtering, search, favorites, one-click prompt copying, and JSON export. Images were rendered on a 3090 (16MP, 6 steps, no CFG, 35s each).
More from Multimodal
- Fable generates 12 full videos in one shot with sound, voices and custom songs — RileyRalmuto · 2026-11-12
- Fei-Fei Li shows CoRL camera-control leaderboard: Alias tops Gemini at 1478 — DominiqueCAPaul · 2026-11-12
- Inflect v3 released: 34M-parameter local TTS with six voices and event tags — b111ue · 2026-11-12
- Coded cartoon show made with Fable 5.2 hints at interactive LLM entertainment — mickeyxfriedman · 2026-11-12
- Frontier Models Fail Hilariously at Drawing a Satyr, While Image-Gen Succeeds — teortaxesTex · 2026-11-12
- Mariko Mori used Claude to code generative animation for her retrospective "All That Shines" — claudeai · 2026-11-12