tiktok-gemini-video-workflow

Submit a product image and the viral video through a Feishu form linked to the workflow's Base. The Skill breaks down the reference frame by frame, generates a 9:16 storyboard and detailed English video prompts, then submits them to Gemini or Flow and returns the finished clips for review.

It reads live quota status from the account's settings page rather than assuming fixed limits, rotates across all enabled Gemini accounts first, then falls back to Flow credits. When everything is exhausted it records the recovery text and waits or schedules a resume only with user authorization.

Yes. You register each account with an alias, Chrome profile name, priority, and switching rules in a separate quota table. The workflow polls accounts by priority and last-used time, keeps the same account sticky per task, and logs every switch with reasons.

Videos up to 15 seconds use one segment and up to 30 seconds use two. The second segment must start from the real extracted final frame of the first segment's downloaded video, combined with the first storyboard and product images to lock continuity.

Yes, it uses the open-source gemini-watermark-remover tool, which reconstructs pixels altered by the visible star watermark. It explicitly forbids delogo, blur, mosaic, or cropping as substitutes, and verifies removal by inspecting extracted frames.

The workflow never relies on the video model to render text hooks. Exact source captions are recorded in a post-production list and added afterward with ffmpeg drawtext or overlay layers, preserving original wording, position, style, and timing before the video can be marked ready for review.