We’re using Veo 3.1 for a family-oriented video commercial. One scene shows a grandfather and grandson walking on a forest path — standard wholesome advertising content.
The problem: Image-to-video generation blocks any anchor image containing a child, even though:
The anchor image was generated by Google’s own Gemini 3 Pro (gemini-3-pro-image-preview) and passed all safety filters
- person_generation is set to allow_all
- safety_setting is set to block_only_high
Errors received:
With anchor image:
The input image contains content that has been blocked by your current safety settings for person/face generation. Support codes: 17301594
With text-only prompt (no image, all person references removed): The prompt could not be submitted. This prompt contains sensitive words that violate Google’s Responsible AI practices. Support codes: 58061214
What we tried:
- Removing all references to children/minors from the prompt — still blocked by the anchor image
- Stripping the prompt to only camera movement and environment — still blocked
- Text-to-video with no anchor — succeeds but produces different character each time, breaking continuity across shots
- Both Vertex AI and Gemini API backends — same result
The core inconsistency: Gemini’s image model generates these images without issue, but Veo refuses to animate them. Google’s own image output is unusable in Google’s own video pipeline.
Request: Is there a path to allowlisting for commercial production involving family scenes?
The person_generation: allow_all setting implies this should be supported.
Support codes: 58061214, 17301594
Cross posting from https://discuss.ai.google.dev/