Reddit tests debunk refmod hype: zero speed gain, possibly a Discord circle promoting friends' projects
dtdisapointingresult · reddit · 2026-11-12
A Reddit user ran benchmarks on the hyped video-gen refmod workflow and found no benefit. Warning signs were visible earlier: no benchmarks or workflows in the README, obvious Claude-slop docs, and development stalled two months ago. Timings: standard ref2va workflow (2 image + 1 audio refs, 0.4MP, 4 steps) at 173s, image=max at 181s, and the refmod-adjusted workflow with precomputed safetensors also 181s — zero speed gain, on top of community reports of degraded references bleeding into each other. The author suspects a small Discord friend group promoting each other's projects regardless of value.
More from Multimodal
- ByteDance launches Seedance 2.5 Flash: nearly 2x faster, 34% cheaper at $0.152/sec — aziz4ai · 2026-11-12
- AI music video blends Suno V6 audio with Dreamina and Seedance visuals — amplifiedamp · 2026-11-12
- AI character design shown in both 2D illustration and 3D render, fans asked to pick — azed_ai · 2026-11-12
- One-line prompt challenge: Fable 5.2 told to make the worst possible video with its full powers — FinanceYF5 · 2026-11-12
- MiniMax spotlights user-made dark fantasy art generated with H3-Image — Hailuo_AI · 2026-11-12
- Fable 5.2 recreates a code-only generated video in Blender, this time as a loop — FinanceYF5 · 2026-11-12