Baseline 20
Original accepted-20 run, with the current evaluator applied to the 17 prompt-valid cases.
Round 1 · Prompt A/B
Direct vs minimal six-section, matched seeds, with final soundtrack delivery where applicable.
Round 2 · Matched Seeds
Compact contract vs minimal six-section; deterministic silence/source soundtrack.
Round 3 · Official Format A/B
Five baseline-success combinations with matched seeds: direct requests vs official six-section prompts. Gemini 3.7 prompt QA approved all five pairs.
Promotion 20 · Matched A/B
Twenty semantically accepted real-workload combinations. Direct and Context-IR-style prompts use identical references, duration, seed, and optimized 4xGB200 runtime.
Repair 15 · Base-Compatible
Fifteen coherent real-workload combinations, three each for A1, A3, A4, B3, and D4. Direct prompts use only FastVideo-bound media labels; unsupported Subject labels are removed.
Audio Repair 5 · No Invented Speech
Five A3/B3 retries with the same reference workloads but an explicit non-verbal contract: mouths at rest, no spoken dialogue, and ambient room tone only.
Audio Repair · Silent Control
The first failed B3 output with its generated audio removed. This isolates speech corruption from visual motion-transfer and source-leakage failures.
Evidence 22 · Fresh Sources, Two Seeds
Eleven fresh, non-overlapping source combinations: five A1, five A4, and one speech-safe A3. Each direct base-native prompt runs at two seeds to measure stability.
10k Distribution v1.4
Deterministic task allocation revised from Pilot-80 evidence. Actual-media semantic acceptance remains a separate stage.
Pilot 80 · Core Distribution
Eighty semantically accepted source combinations using direct, base-native prompts. Raw H3 and postprocessed delivery are reported separately.
Exact Dialogue Stability 8
Four semantically accepted A1/A3 workloads, each repeated at two new seeds. This tests whether explicit short dialogue is more stable than requesting silence.