AI Paper Atlas

Research map / Video / Video generation

Meta Inverse Prompting for Video Generative Models

Yerin Jung, Jongheon Jeong

NeurIPS 2026 · Video › Video generation · approach: Compositional text-to-video diffusion · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes Meta Inverse Prompting, a method for recovering controllable textual conditions from generated videos to invert the usual condition-to-video pipeline.

Problem

Video generative workflows (text-to-video, image-to-video, reference-guided synthesis) are one-directional, offering no mechanism to recover controllable textual conditions from a video.

Approach

Introduces an inverse prompting approach that recovers textual conditions from videos, framed as a meta-level inversion of video generation pipelines; details beyond the abstract are unavailable.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add meta-update (meta-learned initialization) to soft-prompt optimization (ablation)MVIP · MSR-VTTViCLIP V-V: 0.8314 -> 0.8434Improved
use Best-of-N (N=4) candidate selection in inner loop (ablation)MVIP · MSR-VTTDINOv2: 0.6174 -> 0.6516Improved
initialize soft tokens from natural instruction embeddings rather than random (ablation)MVIP · Held-out validation subsetViCLIP V-V: 0.9064 -> 0.9193Improved
increase soft prompt length from 4 to 16 tokens (ablation)MVIP · Held-out validation subsetViCLIP V-V: 0.9129 -> 0.9193Improved
increase soft prompt length from 4 to 8 tokens (ablation)MVIP · Held-out validation subsetViCLIP V-V: 0.9129 -> 0.9045Hurt
increase inner-loop steps K from 1 to 3 during meta-training (ablation)MVIP · Held-out validation subsetViCLIP V-V: 0.9193 -> 0.9174No clear effect
add one test-time adaptation step (K=1) (ablation)MVIP · Real videosViCLIP V-V: 0.8328 -> 0.8392Improved
add a second test-time adaptation step (K=2) (ablation)MVIP · Real videosViCLIP V-V: 0.8392 -> 0.8322Hurt
add iterative ARPO-style refinement to captioningCaptioning (Qwen2.5-VL) · Real videosViCLIP V-V: 0.7977 -> 0.8047Improved
use stronger VLM captioner (Qwen3.5-9B instead of Qwen2.5-VL-7B)Captioning baseline · MSR-VTTViCLIP V-V: 0.8021 -> 0.8312Improved
add a more detailed captioning instructionCaptioning (Qwen3.5-9B) · MSR-VTTViCLIP V-V: 0.8312 -> 0.8123Hurt
transfer the meta-prompt to a different generator (CogVideoX) without retrainingMVIP · Wan2.2 -> CogVideoXViCLIP V-V: 0.8222 -> 0.8709Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy