Cinematic Product Advertisement — Seedance 2.0
A one click workflow to create luxurious cinematic ad videos. Upload images of your product, length of the video, number of unique cuts/shots, and add custom instructions (optional). An LLM processes all this and creates a structured prompt for Seedance 2.0 to generate consistent, on brand product video.
简介
This workflow, Cinematic Product Advertisement — Seedance 2.0, builds multi-shot, cinematic product ads from a single, time-coded prompt. You provide a hero/environment image and product detail references via LoadImage nodes, set the total length and number of cuts using PrimitiveInt nodes, and optionally add a creative brief. GeminiNodeV2 then synthesizes a structured, time-cued shot plan that Seedance 2.0 consumes as one prompt. The ByteDance2ReferenceNode conditions the generator with your hero and detail images, helping maintain brand consistency and product fidelity across every shot.
你需要提供
- 正在读取工作流
在 ComfyUI 中运行。 此工作流无法在本页面运行。你可以在 Comfy Cloud 中打开（已预先加载所需内容），或下载后在自己的 ComfyUI 中使用。
在 Cloud 中试用常见问答
The hero image defines the persistent environment/look for every shot. Detail images focus the generator on specific product features (logos, textures, materials). The ByteDance2ReferenceNode conditions Seedance 2.0 with both so the video stays on-brand and product-accurate throughout.
Edit the time-coded segments in the Edit Prompt? group’s PrimitiveStringMultiline field. For each time range, describe the camera move, product position, lighting, and any transitions. Ensure your segments cover the total duration and reflect the number of cuts you set.
Yes. Paste the prompt into the Edit Prompt? PrimitiveStringMultiline, then bypass the LLM Prompt Generation group. This locks the plan for reproducible renders and faster iteration.
Use a clear, wide hero/environment photo, high-quality product detail images, and concise time-coded directives. Keep the first draft short (fewer cuts, shorter duration), then lengthen or add shots as needed. If you see drift, tighten language in each time segment and avoid conflicting descriptors.