This ComfyUI workflow builds a cinematic, multi-shot product advertisement video from up to three reference images. In the Inputs group, you supply Image 1 as the hero/environment reference (the scene this image defines is maintained across all shots) and use Images 2–3 for product detail references. Duration and number of unique shots are controlled with PrimitiveInt nodes, while optional brand/style guidance goes into a PrimitiveStringMultiline input. The LLM Prompt Generation group uses GeminiNodeV2 to transform your inputs into a single, time-cued prompt that MiniMax H3 understands (for example: 0–3s Shot 1, 3–6s Shot 2, etc.). The structured prompt, plus your references, then drives the MiniMax H3 node to render a consistent, on-brand video.

What makes this workflow especially efficient is its one-prompt multishot design. MiniMax H3 expects a single prompt containing all shots and their time windows, so the GeminiNodeV2 node formats a clean shot list with camera moves, transitions, and product highlights. You can preview and verify the shot script using PreviewAny and MarkdownNote. Once satisfied, paste the final script into the Edit Prompt? group to "lock" it and bypass the LLM step for fast, repeatable iterations. The Generate and Save Video group then renders the ad and writes it to disk via SaveVideo, giving you a repeatable pipeline for quickly producing consistent product spots.