@starreel/mcp 0.1.173 → 0.1.175

package/SKILL.md

@@ -547,11 +547,17 @@ content that will be rejected.
    **Video engine**: videos use a drama-level engine, set via `create_drama` /
    `update_project_settings` field `video_engine`. Four options — surface the
    choice to the customer with the price gaps and let them decide:
-   `seedance-2.5` (default for anime / stylized dramas; full capability: frame
-   chain / scene groups / in-place edit / extend / reference anchors; 720p ≈
-   212 pts/s; **realistic live-action subjects are not available on it for
-   now**); if `video_engine` is omitted the platform picks by style —
-   realistic or unspecified → `hailuo-3`, anime / stylized → `seedance-2.5`;
+   `seedance-2.5` (full capability: frame chain / scene groups / in-place
+   edit / extend / reference anchors; 720p ≈ 212 pts/s); if `video_engine` is
+   omitted the platform only applies a style-based fallback stamp — anime /
+   stylized → `seedance-2.5`; realistic or unspecified → `seedance-2.5` while
+   its realistic switch is open, otherwise `hailuo-3` — which does not check
+   the menu's per-subject availability, so **pass the menu-selected
+   `video_engine` explicitly**.
+   **Which engine is available for which subject class changes at runtime** —
+   always read `availability` from `get_production_menu`; an unavailable entry
+   carries `customer_note` and `expected_until`, relay them to the customer as
+   is. Never assume from this document that a given model is unavailable;
    `hailuo-3` (MiniMax H3; ≈1/3 cost at 70 pts/s for 720p; native dialogue &
    SFX baked in; up to 2K; ~6 min per shot; in-place edit — high fidelity —
    and episode extend supported, input video billed per second on top);
@@ -577,9 +583,10 @@ content that will be rejected.
    reference image's raised-arm pose" produced exactly that raised arm at the
    opening, overriding the source video's motion.
    **How to choose (guide the customer proactively)**: ① realistic live-action
-   dramas → `hailuo-3` (~1/3 cost, up to 2K, slower at ~6 min/shot;
-   `seedance-2.5` does not take realistic live-action subjects for now — it
-   stays available for anime / stylized work); ② stylized / animated /
+   dramas → `seedance-2.5` (best instruction-following and face detail), or
+   `hailuo-3` to cut cost to ~1/3 (up to 2K, slower at ~6 min/shot) — only
+   among the engines `get_production_menu` lists as available for
+   `realistic_human`; ② stylized / animated /
    3D-cartoon dramas, empty shots and product shots → `wan3.0` (~40% cost,
    2s-minimum billing), or `wan3.0-prime` when delivery speed matters;
    ③ NEVER pick wan3.0/prime for realistic live-action — WAN's output
@@ -1131,9 +1138,10 @@ Four distinct causes; they need **opposite** fixes, so identify the family first
    generated after the Aug-30 native-audio fix, so this is the engine itself).
    `wan3.0` is **not measured**: its 26 sampled shots (46%) all predate that
    same fix, and only 2 shots exist after it — treat WAN as unknown here, not
-   as bad. For dialogue-heavy stylized work prefer `seedance-2.5`; on
-   realistic dramas (`hailuo-3` for now) run `scan_dialogue_coverage` after
-   generating and repair short lines with `repair_episode_dialogue`.
+   as bad. For dialogue-heavy work prefer `seedance-2.5` when the menu lists it
+   as available for the drama's subject class; on `hailuo-3` run
+   `scan_dialogue_coverage` after generating and repair short lines with
+   `repair_episode_dialogue`.
 4. **The transition ate the tail.** A cross-fade pulls the *next* shot's start
    backwards, covering the end of the current one. The platform now shortens or
    drops that transition automatically so it can never cover a spoken word, so
@@ -1315,8 +1323,8 @@ start in the middle.
 
 Pick the engine **before** generating video — switching does not retroactively
 fix shots already made. For anything with dialogue and a story to follow, use
-`hailuo-3` for realistic live-action (realistic subjects are not available on
-`seedance-2.5` for now) or `seedance-2.5` for anime / stylized. Reach for `wan3.0` when the shots
+`seedance-2.5` (or `hailuo-3` to cut cost) — whichever `get_production_menu`
+lists as available for the drama's subject class. Reach for `wan3.0` when the shots
 are short and stylised — animation, 3D cartoon, empty/scenery shots, product
 shots — where an extra angle change inside a shot does not read as a mistake.
 Two hard limits on WAN: realistic human faces at 720p+ are rejected outright by

package/dist/i18n/instructions.en.md

@@ -1,28 +1,28 @@
-StarReel = a prepaid AI short-drama production line (script → AI rewrite → assets → storyboards → shot frames → video → audio → final cut .mp4), 120+ tools, all served by this server.
+StarReel = prepaid AI short-drama production pipeline (script → AI rewrite → assets → storyboards → shot frames → video → audio → final cut .mp4), 120+ tools, all via this server.
 
-★ Step one is ALWAYS to judge "what material does the customer have" — it decides the entry point, and rework from a wrong entry costs real credits (full table and key points: get_capabilities_guide):
-· A novel / story outline / synopsis (not yet in script form) → create_drama → set_script → rewrite_script → review_script
-· A finished script (scene headings + dialogue lines) → set_script → rewrite_script → get_script → review_script
-· They want to rewrite on another AI platform themselves / complain "your AI changes too much" → get_script_format_spec → let the external platform rewrite → check_script_format (errors down to zero) → exit A: adopt_external_script (with annotations · adopted as-is · free) or exit B: set_script + rewrite_script (story layer only · platform adds annotations)
-· A finished storyboard table (seconds / shot size / camera move per shot; typical for pilots or sheets handed to TV stations and brands) → get_storyboard_table_spec → tidy it with an external tool against the template → check_storyboard_table (errors down to zero) → import_storyboard_table (auto_complete by default) → get_autofill_status until done → review_storyboards
-· Structured data (customer's own tools / spreadsheet export / third-party AI emitting JSON, to build the episode's storyboards in one go; characters / scenes / props are still extracted by extract_assets) → get_bulk_import_spec → check_bulk_import (errors down to zero) → bulk_import_storyboards (auto_complete by default) → get_autofill_status until done → review_storyboards
-· Their own character sheets / scene plates / prop sheets / shot frames (the customer's real material) → upload_image · set_character_portrait · upload_scene_image · upload_prop_sheet · upload_shot_frame (customer-owned material only; to change the picture use generate_shot_frame) · upload_shot_footage (a whole live-action / screen-recorded clip as a shot's final footage)
-· Their own voice sample / a specific voice → clone_voice → speak_with_voice (preview) → set_character_voice (native-voice dramas: before generate_videos) / assign_voices
-· No external voice source; they want a voice a character already spoke in the film as the reference → generate_videos (one shot per speaking character first) → set_voice_anchor_from_shot → generate_videos (the remaining shots) → get_pipeline_status and check stale_shots
-· A song + lyrics (MV) → create_drama → set_mv_lyrics → generate_mv_story → generate_mv_script → get_mv
-· Lyrics only, no song (they want an MV) → set_mv_lyrics → suggest_mv_song_styles → quote_mv_song (customer agrees) → generate_mv_song → get_mv_song (preview) → select_mv_song → generate_mv_story → generate_mv_script
-· A product / brand (ads, brand films) → create_drama → add_product → generate_product_sheet → list_products → render_multi_aspect
-· Generated shots / a final cut need changes (dialogue, trimming, splitting, regenerating a shot, switching engines) → scan_dialogue_coverage / scan_intra_shot_cuts to diagnose first → update_shot / replace_shot_dialogue / repair_episode_dialogue / split_shot / trim_shot / regenerate_shot_video / edit_video_shot → rerender_episode
-· An existing video (own footage or a generated shot) where they want to keep the motion but replace the character / wardrobe entirely (video repaint) → get_repaint_capabilities → upload_shot_footage (own video) → quote_repaint_shot_video / quote_reskin_shot_video → repaint_shot_video / reskin_shot_video → get_repaint_status
-· They want to edit themselves and need a per-shot handoff pack (clean footage / dialogue track / SFX / music / subtitles) → export_handoff_pack → get_handoff_toolchain → save_handoff_toolchain
-· Multilingual release → translate_subtitles → update_project_settings
+★The first step is always to determine “what materials the client already has” — this decides the entry point; choosing the wrong entry causes real paid rework (full table and key points: get_capabilities_guide):
+· Novel / story outline / synopsis (not yet in script form) → create_drama → set_script → rewrite_script → review_script
+· Already-written script (with scene headings + dialogue-line structure) → set_script → rewrite_script → get_script → review_script
+· Wants to use another AI platform to rewrite / says “your AI changed too much” → get_script_format_spec → rewrite on external platform → check_script_format(errors cleared) → Exit A adopt_external_script(with annotations · direct adoption · free) or Exit B set_script + rewrite_script(story layer only · platform adds annotations)
+· Completed final storyboard table (per-shot duration / shot size / camera movement; common for samples, TV stations, or brands) → get_storyboard_table_spec → organize with external tool using template → check_storyboard_table(errors cleared) → import_storyboard_table(default auto_complete) → get_autofill_status until done → review_storyboards
+· Structured data (client’s own tool / spreadsheet export / third-party AI outputs JSON directly; wants to build this episode’s storyboards in one pass; characters/scenes/props still extracted by extract_assets) → get_bulk_import_spec → check_bulk_import(errors cleared) → bulk_import_storyboards(default auto_complete) → get_autofill_status until done → review_storyboards
+· Own character sheets / scene images / prop images / shot frames (client real assets) → upload_image · set_character_portrait · upload_scene_image · upload_prop_sheet · upload_shot_frame(client-owned assets only; to alter the image use generate_shot_frame) · upload_shot_footage(full live-action/screen-recording clip as one shot’s final video)
+· Own voice sample / specified voice → clone_voice → speak_with_voice(test listen) → set_character_voice(original-voice drama: before generate_videos) / assign_voices
+· No external audio source, wants to use a voice already spoken by a character in the video as baseline → generate_videos(first generate one shot for each speaking character) → set_voice_anchor_from_shot → generate_videos(remaining shots) → get_pipeline_status check stale_shots
+· Song + lyrics (MV) → create_drama → set_mv_lyrics → generate_mv_story → generate_mv_script → get_mv
+· Lyrics only, no song (wants MV) → set_mv_lyrics → suggest_mv_song_styles → quote_mv_song(user agrees) → generate_mv_song → get_mv_song(test listen) → select_mv_song → generate_mv_story → generate_mv_script
+· Product / brand (ad, brand microfilm) → create_drama → add_product → generate_product_sheet → list_products → render_multi_aspect
+· Generated shots / final cut need changes (change lines, trim, split shot, regenerate a shot, switch engine) → scan_dialogue_coverage / scan_intra_shot_cuts diagnose first → update_shot / replace_shot_dialogue / repair_episode_dialogue / split_shot / trim_shot / regenerate_shot_video / edit_video_shot → rerender_episode
+· Existing video segment (own asset or generated shot), wants to keep action but replace character / outfit overall (video repaint) → get_repaint_capabilities → upload_shot_footage(own video) → quote_repaint_shot_video / quote_reskin_shot_video → repaint_shot_video / reskin_shot_video → get_repaint_status
+· Wants to edit themselves: needs per-shot asset pack (raw video / dialogue tracks / SFX / music / subtitles) → export_handoff_pack → get_handoff_toolchain → save_handoff_toolchain
+· Multilingual distribution → translate_subtitles → update_project_settings
 
-Pipeline order (no skipping): create_drama (set setting_brief / aspect ratio / video_engine / image_model / consistency anchors at creation — all free) → set_script → rewrite_script → ★review_script → extract_assets → quote/generate_storyboards → ★review_storyboards → character sheets + design sheets / scene plates / prop sheets / color script → ★review_assets → run_precheck → quote/generate_frames (first frames only by default) → tail_frame_plan (free · which shots need a separate last frame) → quote/generate_frames (frame_type=last_frame) → chain_frames (seams · dry_run:true first to see the plan, run it after the customer confirms) → get_jobs (wait until its reframe image jobs finish; failed ones show error_msg) → ★review_frames → quote/generate_videos → audio → ★review_videos → compose_episode → get_final_cut → review_final_cut. Track progress with get_pipeline_status.
-Five free hard gates (skipping → 400): review_script (before extract_assets / storyboards) · review_storyboards (before frames, incl. per-shot generate_shot_frame) · review_assets (before frames; pass its token as asset_review_token) · review_frames (before videos, incl. per-shot regenerate_shot_video) · review_videos (before compose_episode); pass the review_token to downstream paid tools (batch and per-shot alike) and relay every finding to the customer verbatim.
-★After every review you must also do your own QC: open each item in the returned agent_qc materials yourself, answer its checklist, and submit_qc_report; on gated dramas the token is unusable until the report is in.
-Billing discipline: prepaid, no overdraft; for large steps call quote_* → tell the customer estimated_points as-is → only after they agree call generate_* (quote_id), never confirm on their behalf; image quotes give estimated_points (upper bound — prepare balance for it) and typical_points (usual spend) — say both; text steps are billed post-hoc by tokens; on 402 stop and let the customer top up, do not retry.
-Long tasks are async: generate_* returns immediately; poll with get_pipeline_status / get_storyboards / get_run_status; an image in pending is still generating — do not call again (double billing).
-After a successful rewrite only spot-edit with edit_rewritten_script, do not rerun rewrite_script; the single source of truth for character appearance is the character profile (update_character) — never put it into visual_lock / art_bible.
-Choosing an image model / video engine, or the customer asks "which model is better / how to save" → get_capabilities_guide (section=model_guide): measured strengths and weaknesses, prices, decision steps and recommended combos; bring it up proactively when creating the drama.
-Action / fight / skill-showcase films → get_capabilities_guide (section=action_film): the storyboard-table fast path, first frame frozen on the wind-up, grouping and rhythm cutting.
-Unsure which tool to use, or the customer asks "what can you do" → call get_capabilities_guide first (free, local, no network).
+Pipeline order (do not skip steps): create_drama(create drama and set setting_brief/aspect ratio/video_engine/image_model/consistency anchors, all free) → set_script → rewrite_script → ★review_script → extract_assets → quote/generate_storyboards → ★review_storyboards → character sheets + turnarounds (generate_character_sheets) / scene images / prop images / color script → ★review_assets → run_precheck → quote/generate_frames(default only first frame) → tail_frame_plan(which shots need independent last frames) → quote/generate_frames(last_frame) → chain_frames (dry_run first, run it after confirmation) → get_jobs (wait for its reframe image jobs to finish) → ★review_frames → quote/generate_videos → audio → ★review_videos → compose_episode → get_final_cut → review_final_cut. Use get_pipeline_status to check progress.
+Five free hard gates (skip → 400): review_script(before extract_assets/storyboards) · review_storyboards(before frame generation via generate_frames, including single-shot generate_shot_frame) · review_assets(after character/scene/prop images are generated, before frame generation via generate_frames / generate_shot_frame; pass its token as asset_review_token) · review_frames(before video generation, including single-shot regenerate_shot_video) · review_videos(before compose_episode); pass review_token to downstream paid tools (both batch and single-shot tools require it); report findings to the client verbatim item by item.
+★After each review layer you must also personally QC: inspect assets item by item according to returned agent_qc, then submit_qc_report; gray-release dramas cannot proceed without the report.
+Billing discipline: prepaid, no overdraft; for large steps quote_* → tell client estimated_points exactly as returned → after client agrees, generate_*(quote_id); never confirm on your own; image-generation quotes provide estimated_points(upper bound, prepare balance by this) and typical_points(usual spend), state both; text steps are postpaid by token; on 402 stop and ask client to recharge, do not retry.
+Long tasks are async: generate_* returns immediately; poll with get_pipeline_status / get_storyboards / get_run_status; image pending = still generating, do not call again (duplicates charge credits).
+After rewrite succeeds, only use edit_rewritten_script for targeted edits; do not rerun rewrite_script; the single source of truth for character appearance is the character profile (update_character), do not write it into visual_lock/art_bible.
+Choosing image model/video engine, or client asks “which model is better / how to save more” → get_capabilities_guide(section=model_guide): measured pros/cons, prices, decision steps, and recommended combos for each model; proactively explain and decide with client when creating the drama.
+For action / fight / skill demo videos → get_capabilities_guide(section=action_film): storyboard-table fast path, first frame frozen at wind-up pose, grouping and rhythm trimming.
+If unsure which tool to use, or client asks “what can you do” → call get_capabilities_guide first (free, local, no internet).