FramePack-U: Training-free constant-memory frame packing boosts long-video VLM benchmarks by up to 12.9%
UNC-ChapelHill · hf · 2026-10-14
Researchers from UNC-Chapel Hill, Renmin University and UW propose FramePack-U, a training-free frame packing framework that compresses arbitrarily long videos into a fixed-size visual representation at constant memory cost.
- Scores each frame by semantic relevance to the text prompt, then selects frames in a diversity-driven manner
- Packs selected frames at varying resolutions into a fixed-length context window
- Requires no fine-tuning or architectural changes; plugs into off-the-shelf VLMs at inference time
On Qwen3.5-35B-A3B with a 128-frame budget, FramePack-U gains +12.9% on LongVideoBench, +10.6% on MLVU and +9.7% on Video-MME over the official model, consistently beating token-reduction and keyframe-selection baselines across model families and scales. Code and data are open-sourced.
More from Multimodal
- Dev builds Grok Bot chaining Suno, Midjourney and Opus to autonomously make music videos — TinfoilTricorn · 2026-10-14
- This Midjourney --sref 1330146212 nails a moody 90s film-look aesthetic — umesh_ai · 2026-10-14
- Redditor Has Fun Testing MiniMax's Text-to-Video Generation — NapoletanoPizzaiolo · 2026-10-14
- Koboldcpp v1.123 adds support for Qwen Image 2.1 — Fcking_Chuck · 2026-10-14
- AI Horror Short Asks a Billion AI Minds Whether God Exists — JoelXGGGG · 2026-10-14
- Midjourney v8.2 style demo: dreamy red-blue motion-blur photography with sref 4075477523 — azed_ai · 2026-10-14