Source: raw/This_FREE_Claude_Skill_Powers_My_Entire_Seedance_2.0_Workflow.md — YouTube tutorial (creator unattributed in transcript head), 2026-05-28 fetch.

A free, single Claude Skill that prompts three different generation models — Nano Banana Pro, GPT Image 2, and Seedance 2 — in their own natural prompt language for three distinct AI-filmmaking jobs. Distinct positioning from Adil’s Higgsfield-MCP Content Factory: that skill is marketing-UGC-focused via Higgsfield MCP; this skill is cinematic-short-film-focused via LTX Studio, with no MCP — pure prompt-engineering wrapper. Same skill-as-distribution-unit pattern; different use case + different creative engine.

Key Takeaways

  • One skill, three prompt languages. The skill ships templates for: (1) character sheet (prompt for Nano Banana Pro / GPT Image 2 to lock character consistency), (2) 3×3 cinematic storyboard grid (prompt with per-panel choreography + production notes), (3) Seedance 2 shot (continuous-shot prompt OR timestamp-based prompt with dialogue + storyline). Invoked by natural language — “create a character sheet prompt for attached character”, “create a storyboard grid for a fight scene”, “create a Seedance 2 prompt for attached cinematic grid”.
  • The bottleneck the skill solves: three jobs, three prompt languages. Most operators use one prompt formula for everything. Character sheet ≠ 3×3 grid ≠ Seedance 2 shot. The skill recognizes which job is being asked and consults the right template.
  • Creative engine: LTX Studio (not Higgsfield). The tutorial uses LTX Canvas for inspiration storage, LTX Studio for actual generation across all three models in one place. LTX Studio is positioned as “all-in-one with all the latest image and video generation models” — and the skill is engine-agnostic on the prompt-output side (Claude writes the prompt; you paste it into LTX).
  • Workflow stages:
    1. Drop an image reference into LTX Canvas as inspiration.
    2. Attach to Claude, invoke the skill (“create a character sheet prompt for attached character”).
    3. Paste Claude’s output prompt back into LTX Studio with the image reference → generate character sheet.
    4. With character sheet in hand, ask Claude for a 3×3 cinematic storyboard prompt for the scene narrative.
    5. Paste storyboard prompt into LTX, generate the 3×3 grid using GPT Image 2 (with character refs).
    6. Attach the grid to Claude, ask for a Seedance 2 prompt. Skill writes prompt referencing “image one as cinematic grid”, “use as main visual and motion reference”, 10-15 second sequence, “read storyboard panels as sequential shots”.
    7. Paste Seedance 2 prompt into LTX, generate the final video.
  • Two Seedance 2 prompting paths: (a) storyboard-grid-driven (continuous-shot fight scene), (b) timestamp-based with character sheets and dialogue (scene with multiple lines of dialogue, multiple character references). The skill recognizes both modes from the natural-language request.
  • The grid-as-reference trick. Seedance 2 accepts the 3×3 storyboard grid as a single image reference and reads the panels as sequential shots within a 10-15s clip. This is the load-bearing technique that turns a static storyboard into directable motion.
  • Skill install path (standard custom-skill). Download zip from video description → Claude → Customize → Skills → “+Add skill” → Create skill → Upload skill → drag and drop the zip → ensure enabled (check mark). Same path as any other custom Claude Skill.
  • Distribution model. Single zip download from a YouTube video description. Free. No login, no MCP, no recurring auth — the skill is pure prompt-engineering templates + a SKILL.md.

How it compares to Adil’s Higgsfield-MCP Content Factory

Both ship a custom Claude Skill for AI video workflows. Distinct use cases:

DimensionThis skill (Seedance + LTX)Adil — Higgsfield MCP Content Factory
Target outputCinematic short filmsMarketing UGC ads
Creative engineLTX StudioHiggsfield Marketing Studio
ConnectorNone (prompt-only)Higgsfield MCP
Trigger modelNano Banana Pro / GPT Image 2 / Seedance 2Higgsfield + GPT Image 2
Skill primitivescharacter sheet, 3×3 grid, Seedance 2 promptresearch, content plan, generate, meta-ads upload
Permission/cost gatingOperator controls each paste-and-generatePer-batch permission gates inside the skill
Final publishManual uploadMeta Ads upload as final stage
DistributionSingle zip from video descriptionSingle zip from video description

These are sister skills, not competitors. They solve adjacent problems with the same “Claude Skill as the load-bearing primitive” pattern.

Try It

  1. Watch the source video to follow the exact LTX workflow: Fark3A0ACzM.
  2. Download the skill zip from the video description.
  3. Install: Claude → Customize → Skills → “+Add skill” → Create skill → Upload skill → drop zip → confirm enabled.
  4. Run the smallest possible flow first. Just “create a character sheet prompt for [attached image]” → paste into LTX → confirm the skill is loading. Then add the 3×3 grid step. Then the Seedance prompt step.
  5. Try BOTH Seedance 2 prompting paths: (a) grid-as-image-reference, (b) timestamp-based with character sheets + dialogue. They produce structurally different outputs; pick the one that fits the scene type.
  6. Use the grid-as-reference trick for any continuous-shot scene (fights, chases, dance) — the panels read as sequential shots within Seedance 2’s 10-15s window.

Open Questions

  • Creator attribution. Video author not named in the transcript head; YouTube ID Fark3A0ACzM. Worth a follow-up identification before treating the skill as load-bearing — same flag that caused this article to be deferred from the 2026-05-28 16:50 compile pass.
  • Skill source code visibility. Distributed as a download zip rather than open source. Worth inspecting the SKILL.md and template prompts before adopting widely.
  • Single-skill vs. plugin-bundle. This is one skill with three template branches. Adil’s Content Factory is also one skill with four template branches. Whether multi-job skills consistently beat plugin bundles for AI-filmmaking workflows is an emerging pattern across the topic.
  • LTX Studio vs. Higgsfield trade-off. This tutorial picks LTX for the unified multi-model surface (Nano Banana Pro + GPT Image 2 + Seedance 2 in one UI). Adil picks Higgsfield via MCP for connector-driven automation. Direct head-to-head on cost-per-finished-second + creative control would clarify which engine each operator should default to.
  • Seedance 2 multi-character consistency at 15s. The transcript demonstrates two-character dialogue (woman + creature) at the LTX/Seedance combo. Whether character consistency holds across longer cuts (>15s sequence requires multiple generations stitched) is not addressed in this video.