SenseTime Seko Video Agent Update
SenseTime's Seko video agent announced a feature and pricing update that expands creative model options and clarifies workflow and cost. Seko now offers Seedream 4.0 (Seedream) and Nano Banana models on a free/trial basis while applying a 50% discount to the Veo3 API. The agent supports end-to-end video creation: storyboarding, shot-level editing, voiceover, background music, lip-sync, subtitles, and clip-length adjustments. Models are categorized by resolution (720p vs. 1080p), with the UI showing which model is used and the points cost per generated video. The workflow requires uploading the primary subject during the creative phase and referencing it to keep frame consistency. These changes lower barriers for personal-IP animation and give users shot-level stylistic control and clearer cost visibility.
Key insights
Promotional pricing and model access
Seko has made Nano Banana and Seedream 4.0 available under free/limited-access terms and reduced the Veo3 API price by 50%, enabling lower-cost experimentation with high-quality video models.
Shot-level creative control and multi-function editing
Seko exposes a page-level interface where users can modify storyboards, dubbing, BGM, lip-sync, subtitles, clip durations, and add or remove shots, enabling end-to-end per-shot adjustments within the agent.
Model grouping for frame consistency
Seedream 4.0, Nano Banana, and Flux.1 Kontext Pro are treated as a consistency image-model group, helping maintain consistent character and scene appearance across storyboard frames while allowing per-shot stylistic edits.
Resolution-based model selection and transparent cost display
Seko separates video models by resolution tiers (720p and 1080p) and lists which model was used along with the total points cost for generated videos so users can predict consumption and budget accordingly.
Recommended creative workflow step
During the creative planning stage Seko requires uploading the primary on-screen subject and referencing it (e.g., using an @-tag) in the scenario so subsequent model-driven storyboard generation preserves the intended performer identity.