deepfake-video-detection-7-8b-open-multimodal-backbones-splitting-analysis CLM-0558For deepfake video detection with 7-8B open multimodal backbones, splitting the analysis into four separately prompted perspective agents (texture, lighting, motion, physics) whose reports a judge agent reconciles beats the same backbone doing chain-of-thought or multi-turn self-questioning over the same four perspectives, and the gain is not explained by extra turns or repeated frame access.
Capability: Getting more from several models than from one
Observed on
Out-of-domain split of one 100K-video face-forgery dataset, 20 held-out generators; Qwen2.5-VL-7B and InternVL3.5-8B; absolute scores remain low (F1 around 30-53)..
Sources
- Validated by: Training-free comparison on the OOD split against Single, CoT, and three multi-turn variants with the same backbone: F1 33.53 vs 13.24 (Single) and 11.29 (CoT) on Qwen2.5-VL-7B; after SFT plus GRPO the full system reaches 69.87 accuracy / 53.28 F1 vs 47.45 F1 for the best single-model baseline (Gemini-2.5-Pro). Per-agent ablation shows all four combined beats any single perspective.. Measured on 7,636 OOD videos (5,716 real, 1,920 fake) from 20 unseen generators. Training-free comparison isolates architecture from training; [truncated]
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For 4-8B multimodal models on long-video QA, RL training on questions authored from short scene segments but answered over the whole video improves benchmark accuracy more than RL on questions authored from the whole video, because segment-authored questions target details a whole-video view overlooks; including the segment timestamp in the training instruction adds a further gain by removing referential ambiguity, and is not needed at inference.Reasoning about time in video · reviewed by AI
- On multi-hop video questions written so no single frame or timestamp lookup suffices, single-turn MLLMs answer poorly (0.07-0.51 accuracy across 12 proprietary and open models), and wrapping the same models in an agentic loop helps only when the loop lets the agent choose what, where and at what granularity to look: frameworks giving adaptive fps/resolution/interval control raised accuracy and trajectory scores for nearly every model, while a fixed-schedule framework with weak perception tools (STAR) fell below the single-turn baseline for all of them.Reasoning about time in video · reviewed by AI
- For VLM-based judging of camera-conditioned world-model videos, splitting a long video pair into action-aligned chunks with structured visual evidence (source image, frame-grid overview, per-action first/last-frame panels) and voting chunk verdicts into video-level preferences agrees with human preferences better than judging the whole video pair at once, because short-lived action evidence is otherwise diluted by the full context.Judging another's work as well as a careful reviewer would · reviewed by AI
- For long-video QA under a fixed frame budget with a frozen 7B multimodal model, organizing frames into a question-agnostic hierarchical temporal tree and searching it coarse-to-fine (answer, then issue a search statement, then expand the relevant branch) raises accuracy over uniform sampling and over agent methods that score frames independently, with the gain largest on the longest-video benchmarks.Reasoning about time in video · reviewed by AI
- For 2B-8B video multimodal models fine-tuned for temporal reasoning, placing a small minority of reversed-video rollouts in each GRPO group (1 reversed out of 7) beats both all-forward GRPO and larger reversed shares (2 or 3 of 7), because more reversed samples dilute the main forward-video answering objective.Reasoning about time in video · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: low. Falsifier as drafted: A multi-turn or CoT prompt that elicits the same four dimension-specific analyses from one agent matching the multi-agent framework's F1 on the same OOD split and backbone, or the multi-agent gain disappearing when judge and observation agents are given matched token budgets.