baseball-trend

baseball-trend

15-second ESPN-style broadcast cutaway of a user, sitting behind home plate at a fake Yankees vs Red Sox ALCS Game 3 game at Fenway Park, with two announcers naming them on air.

Fixed-recipe skill — the prompts below are calibrated. Substitute the username and keep the marked anchors intact.

Cost transparency gate

Before any paid MCP call, call identity_balance({verbose: true}) once. Surface the current balance, recent burn rate, and remaining runway, then gate the run with this exact message:

Estimated cost: about 3,000-5,500 credits (~$30-$55) for the GPT-image-2 broadcast still, one or two Kling v3-omni pro 15s renders (includes one Step 2 corrective retry with a changed payload), and post-flight analyze_media QA. This exceeds $5, so Reply proceed to continue or cancel to stop.

Do not call any paid MCP tool until the user replies proceed. If the user replies cancel, stop without generating. This is the only yes/no gate; after proceed, the pipeline runs end-to-end.

Voice selection note

This skill uses Kling-omni's native broadcast commentary: two male announcers, matching MLB broadcast convention. The identity_voice setting is NOT consumed because this fixed recipe does not use agent-side TTS or custom voice IDs.

If the user wants a female-coded announcer or any custom voice, baseball-trend is the wrong skill. Route them to /pika:podcast with a baseball framing, which has an agent-side voice path and can honor explicit voice choices.

Stage 0 — Intake

If invoked with empty args and no usable prior context, print this menu and stop:

Who should appear in the fake MLB broadcast cutaway? Required:

• Name — exactly as it should appear in the chyron and announcer dialogue
• Reference photo — one front-facing or 3/4 portrait, local path or HTTPS URL

If only one field is missing, ask only for that field. Otherwise ask the two questions below one at a time.

1. Username (required) — used both in the broadcast chyron and in the announcers' commentary, e.g. "Jane Doe". Save as state.username. This replaces every literal ${username} in the prompts below.

2. Reference image (required) — one front-facing 3/4 portrait, good lighting, one face. Resolve to a CDN URL and save as state.reference_image_url:

• https://… URL → use as-is.

• Local path → upload the file with upload_asset, then use the returned public URL.

• Claude Desktop, photo pasted inline → inline pastes don't reach MCP tools yet (Anthropic limitation). Reply with:

  Heads up — pasted images don't reach MCP tools on Claude Desktop yet. Two options:

  • Paste a URL if it's already hosted somewhere — fastest.
  • Attach the image file so I can upload it before generation.

  When a local file arrives: convert it to a public URL with upload_asset and use public_url.

After both answers are in, run the Cost transparency gate, then echo one short confirmation ("Generating behind-home-plate cutaway for {username}…") and start the pipeline. No further yes/no gates after the cost gate — the pipeline runs end-to-end.

Stage 0.5 — Avatar-type probe for the reference image

Before any paid generate_image_edit or generate_reference_video call, run this Avatar-type probe on state.reference_image_url. The baseball-trend recipe turns the subject into a fake broadcast guest, so IP-derived avatars and celebrity/public-figure references are especially likely to fail moderation.

Call analyze_media once:

query: "Classify this image for paid video generation. Is it a photograph of a real human face, an AI-generated realistic portrait, a stylized / illustrated character, or a recognizable trademarked / copyrighted character such as Batman, Pikachu, or Mickey Mouse? Return strict JSON only: { \"avatar_type\": \"real_human\" | \"ai_realistic\" | \"stylized_illustrated\" | \"recognized_ip\", \"recognized_character\": string | null, \"moderation_risk\": \"low\" | \"medium\" | \"high\", \"recommendation\": \"proceed\" | \"warn\" | \"reject\" }. Use null for `recognized_character` when no specific character is recognized; never write \"none\", \"unknown\", or explanatory prose in that field."

Route from the result:

• recognized IP / copyright risk -> STOP only when avatar_type is "recognized_ip", or recognized_character names a specific character (for example "Batman"), or when both moderation_risk is "high" and recommendation is "reject". Treat recognized_character: null, empty string, "none", "unknown", "n/a", and low/medium moderation_risk as not enough to stop by themselves. Run this check before the real/stylized routes. A chibi Batman is still Batman even when avatar_type is stylized / illustrated.
• real human / AI-generated realistic -> proceed normally.
• stylized / illustrated -> proceed with a visible warning that stylized avatars can reduce likeness quality and may be blocked by image/video moderation.
• trademarked / copyrighted -> STOP before generation. Surface this message: Your identity avatar appears to be a trademarked character ([X]). Most video providers will moderate this and refuse to generate. Pass --avatar <real-looking-photo-url> to override, or update your identity avatar at pika.me first. For this skill, ask for a replacement reference photo URL or path if the user did not pass an --avatar <real-looking-photo-url> style override.

Pipeline

Two Pika MCP stages, sequential. Primary engines are locked: gpt-image-2 for the still, kling-v3-omni for the video. The only still-stage fallback is the documented seedream pass below when OpenAI declines the broadcast still with moderation_blocked.

Long-running task_status polling

When any long-running generation call returns a task_id with or without an initial status, including {task_id}, {task_id, status: "queued"}, or an initial queued, running, or processing status, record the task id and start time immediately.

• Call task_status({task_id}) in a tight loop until terminal (completed | failed | cancelled). No manual sleep and no Bash polling; the worker holds each status call open.
• Emit ONE visible progress line every 60s while status is queued, running, or processing: Seedance i2v queued for {N}m {S}s... still processing. Replace the provider/stage label when polling GPT-image-2 or Kling tasks.
• On completed, unwrap the returned result URL and continue.
• On failed or cancelled, surface failure to the user with task_id, status, and the last status message.
• After 15 min total from the original submit, call task_cancel({task_id}) if the task is still non-terminal, then surface failure to the user. If cancel reports the task is already terminal, call status once more and report that terminal result.
• Do not submit a duplicate request while the original task is still queued, running, or processing.

Step 1 — Broadcast still (generate_image_edit)

The chyron + scorebug get baked into the still at frame 0 (load-bearing — when Kling is asked to "pop in" the chyron mid-clip it appears at second 4–5 with a visible flash and breaks the trend; baking it into the first frame makes Kling treat it as pixel-locked burned-in UI).

Call generate_image_edit with:

• provider: gpt-image-2
• images: [state.reference_image_url]
• aspect_ratio: 16:9
• quality: medium (default for speed; high is now exposed but ~2 min/call — use only when fidelity matters)
• output_format: png
• prompt (verbatim, ${username} substituted):

Save the returned URL as state.broadcast_still_url.

Retry budget: Step 1 still generation gets at most 3 total attempts, including the primary still, self-check re-rolls, invalid-image retries, quality downgrades, and the one-shot seedream fallback below. Track state.step1_attempt_count before every paid still call.

OpenAI moderation fallback: if the Step 1 gpt-image-2 call returns moderation_blocked, treat it as a known policy surface for "real person + ESPN-branded live feed + real MLB team context." Do not keep re-rolling the same OpenAI call. Try seedream once with the same images, aspect_ratio, and prompt text, omitting quality and output_format because those are gpt-image-2-only fields.

• If seedream succeeds, save that URL as state.broadcast_still_url, set state.broadcast_still_provider = "seedream_fallback", and continue to the self-check below. The fallback is one-shot: do not call gpt-image-2 again and do not call seedream a second time in this run.

• If seedream also fails or returns a policy/safety block, stop and surface this exact user-facing message:

  OpenAI declined this image. Try using a fresh AI-generated headshot instead of a personal photo, or pick a non-MLB sport variant.

Agent-side self-check before Step 2: the chyron must spell the username correctly and the scorebug must look like real broadcast UI. If either looks wrong after the primary gpt-image-2 still, re-roll Step 1 once per failed self-check while the Step 1 cap has attempts remaining (everything downstream pixel-locks to this frame). If either looks wrong after state.broadcast_still_provider = "seedream_fallback", do not re-roll Step 1; stop and ask the user for a fresh AI-generated headshot or a non-MLB sport variant using the same message above. This is the agent's own check — do not ask the user unless the one-shot fallback path has already failed QA. After either cap is exhausted, stop and ask for a better reference photo or permission to deliver the best attempt; include the best still/video URL and the failing check.

Step 2 — 15s broadcast video (generate_reference_video)

image_types: ["first_frame"] locks state.broadcast_still_url as Kling's literal frame 0, keeping the chyron + scorebug pixel-static for the full 15s.

Call generate_reference_video with:

• provider: kling
• kling_model: kling-v3-omni
• duration: 15
• aspect_ratio: 16:9
• quality_mode: pro
• reference_images: [state.broadcast_still_url]
• image_types: ["first_frame"]
• sound: true
• prompt_adherence: strict (load-bearing — without it the scorebug animates and identity drifts late in the clip)
• negative_prompt (verbatim, load-bearing — without these entries Kling occasionally morphs the scorebug or fades the chyron):

• prompt (verbatim, ${username} substituted everywhere; pre-trimmed to fit Kling's 2500-char cap; chyron-on-frame-0 lock at top):

Save the returned video URL as state.broadcast_video_url. If generation completes asynchronously, follow the MCP tool's returned status handle until the video reaches a terminal state.

Step 2 Kling video generation gets at most 2 total attempts (initial render + one corrective retry for drift, scorebug/chyron movement, or announcer mispronunciation). kling-v3-omni has no seed, and identical Kling payloads can resolve to the same job/asset. Do not submit an identical Kling payload just to seek variation. Before the corrective retry, materially change the payload by using an updated state.broadcast_still_url, restoring missing strict params / negative_prompt entries, or shortening / clarifying the A/B announcer block. Track state.step2_attempt_count. After either cap is exhausted, stop and ask for a better reference photo or permission to deliver the best attempt; include the best still/video URL and the failing check.

Step 3 — Deliver

Return both Pika CDN URLs: the still image URL and the final video URL. If the host client requires local media markers, create the local preview outside this skill after confirming both CDN URLs are reachable.

One-line summary: "Behind-home-plate cutaway for {username} — 15s, 16:9, 1080p, Kling v3-omni, native two-announcer commentary."

Post-flight quality gate

Before declaring success, call analyze_media on state.broadcast_video_url and ask for a structured verdict:

• If announcer_count < 2, treat the result as at least degraded, include audio_warning, and use the one Step 2 corrective retry only after changing the A/B audio payload. Do not submit an identical Kling payload.
• If verdict is clean, return the still URL and final video URL normally.
• If verdict is degraded, return the URLs plus the quality_warning and audio_warning so the user can review before publishing.
• If verdict is catastrophic, do not call the run complete; surface the verdict and re_roll_suggestion instead of declaring success.

Load-bearing phrases (don't strip these)

These are empirical behavior dependencies, not writing style — removing them breaks the recipe:

• In the still prompt: Hardlock: Do not alter their facial structure and maintain their likeness + The subject must match the reference person (without these, identity drifts on the first frame, and everything downstream inherits the drift).
• In the video prompt: Preserve identity strongly + The ESPN scorebug AND the "${username}" lower-third chyron are ALREADY on screen at frame 0 — keep them visible, unchanged, pixel-locked (without these, Kling re-animates the chyron mid-clip).
• The full negative_prompt list — every entry there came from a specific failure mode in prior runs.
• prompt_adherence: "strict" and image_types: ["first_frame"] — see inline notes above.

Engine choice: Kling-only (with one caveat)

Seedance has a two-stage partner_validation_failed 422 gate observed across repeated NBA-sibling runs:

• Input-side (body.image_urls): rejects if the reference contains a recognizable real person.
• Output-side (body.generated_video): rejects AFTER generation if the produced clip contains recognizable-looking faces — and every broadcast cutaway has a crowd full of faces.

The output-side gate is unavoidable for this trend regardless of subject, so Seedance is functionally unusable here. Kling is the engine that works for ordinary user photos.

Kling caveat — recognizable celebrities are blocked too. Kling has its own content-moderation gate that fires on celebrity references. A celebrity-reference prompt plus matching broadcast chyron can fail at submit-time with task_status: failed, task_status_msg: "Failure to pass the risk control system". This is correct behavior — the trend illusion only works with a non-public-figure reference where the chyron name + face are coherent. If a user supplies a celebrity photo, surface the gate to them and ask for a non-celebrity reference instead.

Kling trade-offs: 2500-char prompt cap (recipe above is pre-trimmed). kling-v3-omni has no seed; identical Kling payloads can collapse to the same job/asset, so a corrective retry must materially change the first-frame still, prompt, negative_prompt, or audio wording. Do not submit an identical Kling payload for variation.

Runtime expectations

Typical run time is 4-7 minutes:

Failure modes

What NOT to do

• Don't sport-swap. NBA / NFL / soccer variants → fork this skill; don't parameterize this one.
• Don't add suffixes to the chyron (e.g. " - AI Creator"). Chyron is the username alone — the trend illusion depends on it reading like a real broadcast identifier.
• Don't add post-edits — no add_captions, generate_music, edit_*. Kling burns the scorebug + chyron + native commentary directly; anything added afterward breaks the broadcast illusion.