Image-to-Video Prompting Notes

This note summarizes common prompt patterns across image-to-video tools and how we apply them in this repo’s “Photo to Dance” flow.

Market patterns (high-level)

• The input image defines the subject + composition; the prompt should focus on motion, action, and camera.
• “Image-to-video” prompts generally work better when you describe what changes over time, not what already exists in the image.
• Many tools support negative prompts to reduce artifacts like warping, flicker, extra limbs, or text overlays.

Veo (via Replicate) parameters worth supporting

• prompt
• image (image-to-video, first frame) or reference_images (reference-to-video)
• negative_prompt
• duration (seconds)
• aspect_ratio
• resolution

Recommended prompt structure (Photo to Dance)

Keep it short and motion-first:

• Action: “The child starts dancing…”
• Dance style: “hip-hop / cute wiggle / viral dance…”
• Constraints: “keep identity and clothing consistent; stable background; smooth motion”
• Camera: “locked-off medium shot” (or slow push-in if desired)