MiniMax H3 Unified (Lip-sync, Scene Prompts and Saved-run Edits)

Update Eclipse Nova to 1.0.36 and SmartLLM Nova to 1.1.17What is new in Nova.4<ul><li>Separate opening warmup: give the first lip-sync part a hidden lead-in, such as one second, independently of later reset warmup. The final video retains the complete soundtrack and its timing.</li><li>More transition coverage: Scene Detector now also checks persistent appearance changes across short blends. Review its boundaries; fast pose changes and edits against similar backgrounds can still be missed.</li><li>Multiple lip-sync images: each image interval receives its own scene prompt. Supplied transition times stay in order, and images without a time are spaced evenly through the remaining audio.</li><li>One image, multiple cuts: use manual cut times and separate prompts while reusing one reference. With no cuts and an empty next-part list, a single-image FL2VA scene uses generated continuity at extra render splits.</li><li>Updated scene analysis: Frames/Grid input, editable task instructions, and clearer framing, movement and appearance rules. Scene Description (T2V) still analyzes the source video or images; T2V means the generator has no separate appearance reference.</li><li>Optional video upscale: process the assembled lip-sync or native H3 result before captions and saving. The group uses named variables and can be muted as a whole; downstream groups automatically fall back to the original video and dimensions.</li><li>Stop a stuck LM: the ■ button at the end of Smart LM's Mode bar stops the current workflow and force-stops its managed Ollama/Docker backend. The collapsed bar stays on one row.</li></ul>

MMH3_Unified_N4.json

MMH3_Unified_N41.json

Files with SHA256: c4a569febdb5571502aa8cd689f796242a7c42b72ba573374b4a04313c5cca1e