make-reel

The user wants to create a Poppify reel. Args provided: $ARGUMENTS

Invoke the poppify-build-reel skill and follow its flow:

1. If the user has no apiKey yet, call register() first (optionally { label: "claude" } — there is no source param). Surface the signupBonusUrl so they can claim 50 free seeds before paying.

2. Pick the entry based on what the user provided:

   • Photos (data URLs or http URLs) → start_session_from_photos({ apiKey, photos, goal, audience?, platform? })
   • Topic / brand intent (no photos) → start_session_from_topic({ apiKey, topic, audience, goal, platform }) then refine_concept to pick from the 5 returned concepts. Topic-led sessions start with NO images — every slide's imageUrl is empty. You must give at least one slide an image (search the library or add_slide_image, then update_slides({action:"set_image", slideIndex, imageUrl})) before confirm, or it refunds with topic_led_no_images. One image can serve every beat — set_image the same URL on each slide; captions + motion carry the variety.

3. Review the returned slides + caption + hashtags + CTA with the user. Default to ONE bulk call to apply any edits:

   apply_session_patch({
     sessionId,
     slides: [{index: 0, text: "..."}, ...],
     textColor: "#FFFFFF",         // or any hex
     textAnimation: "phrase_reveal",
     videoEffect: "push_in",      // canonical vocabulary: push_in, pull_out, lateral_pan, vertical_pan, focus_pull, epic_parallax, static
     audioMood: "uplifting",
     audio: { source: "library", assetId: "..." }
   })
   

4. Use search_visual_library and get_music_library BEFORE recommending add_slide_image / add_soundtrack — library matches are free; generation is 5 seeds each. Attach images per-slide via update_slides({action:"set_image", slideIndex, imageUrl}) (or apply_session_patch({slides:[{index, imageUrl}]})) — not the legacy pool. To swap music, apply_session_patch({audio:{source:"library", assetId}}).

5. When ready: call get_result to confirm seed cost — pre-confirm it returns the exact price breakdown — then confirm({ sessionId, apiKey }). Poll get_result every 20–30 seconds. When complete, hand the videoUrl to the user with the note that it's valid ~7 days.

6. Single-image reels: when one image carries the whole reel, keep ONE videoEffect and leave continuousEffect on (default) so the camera makes one continuous move across the slides. Assigning a DIFFERENT effect per slide on a same-image run disables continuous smoothing and produces a visible reset at each cut.

7. Optional Live Motion upgrade (only AFTER the user has seen the cinematic baseline): to animate the subject inside a slide, search_live_library first (cache hits free), else update_slides({action:"set_motion_mode", slideIndex, motionMode:"live", liveAction}) + animate_slide (10 seeds/clip, Veo 3.1 Lite). Recommend at most one live slide, usually the hook. Never apply it before the baseline render. For before/after transformations (endFrameUri), invoke the poppify-live-motion skill first — bridged renders need a composition-locked end frame and an authored journey overridePrompt.

8. Reel length is per-slide — no global knob. Two DRIVERS: text length, or an explicit update_slides({action:"set_duration", slideIndex, duration:N}) (2–15s, works on ANY slide, overrides the text-length formula). Two MEDIA FLOORS that always play in full: attached voiceover audio and a rendered live-motion (Veo) clip — so an 8s morph slide holds all 8s; never pad a caption to keep a clip on screen. Lengthen a slide by writing more text OR set_duration.

9. Optional consistency / before-after (5 seeds/frame, same as an image): for the SAME subject across slides, add_slide_image({referenceAssetId|referenceImageUrl, prompt:"same subject, <new pose/scene>"}) (consistent-frames mode) → set_image, animate normally. For a before/after transformation in ONE slide, add_slide_image the "after" from the same reference, then update_slides({action:"set_motion_mode", ..., endFrameUri:<after url>}) so Veo interpolates start → end. Each frame is one intentional call — not a random batch. Send endFrameUri only for transitions, never for plain micro-motion.