MOSS t2i anti-blur tip: write 'photography', not 'photograph'
Bulky_Chemistry_196 · reddit · 2026-10-21
A practical MOSS text-to-image finding: using the word 'photograph' in prompts tends to produce renders of a physical photo with gloss and blur issues; 'photography' works far better.
The author prefixes realistic-scene prompts with 'RAW photography, candid', since omitting lead words often yields anime output. Pair with the latest kromoss base and base dmd weights for best results.
More from Multimodal
- Fibo2 old-photo restoration and colorization tested on a historic fire scene — Striking-Long-2960 · 2026-10-21
- Full Seedance 2.5 prompt shared for generating ultra-realistic handheld smartphone video — techhalla · 2026-10-21
- GPT Image 2.5 vs Nano Banana 2.1: 502 synthetic rust masks reviewed, 79.7% retained — Gloomy_Recognition_4 · 2026-10-21
- AI-generated 'Doner Kebab Taxi' video goes viral, poster holds the music rights — average1804 · 2026-10-21
- Fibo 2.0 ComfyUI workflows: Edit Turbo is fast and solid at image editing — nomadoor · 2026-10-21
- PVBench: New Benchmark Scores Video Generation Across 46 Physical Plausibility Dimensions — Kaiwei Chen · 2026-10-21