Skip to content

gordonlu/awesome-minimax-h3-skills / community-skills/h3-keyframe-film

h3-keyframe-film

Verified

Write runnable MiniMax H3 keyframe-film prompts — multi-keyframe montages, first-last frame transitions (FL2VA), final-frame landings (L2VA), and I2VA first-frame animation. Use when the user has reference images (keyframes, character art, product shots, final poses) and wants the image identity locked while text drives the motion. Follows base-en.txt exactly, including the keyframe-alignment instruction lines. Pure-text requests are redirected to h3-promo-film; reference-video motion goes to reference-motion-transfer.

H3 Keyframe Film Studio

Use this Skill when the user has images but no video and wants a storyboard-style short from MiniMax H3: "cut through these four keyframes", "transition between these two poses", "land exactly on this final frame". Images lock the identity; text locks the motion.

Core principle: the alignment line decides what the video is anchored to; the timecoded beats decide what happens between anchors.

Prompts follow h3-prompt-writing (base-en.txt). The alignment instruction lines are fixed templates — copy them verbatim.

1. Trigger

Use when: I2VA first-frame requests; FL2VA pose-to-pose transitions; L2VA "end on this exact frame" requests; multi-keyframe montages (up to 4 frames); image + text motion description in any language.

Do not use when: no images at all (→ h3-promo-film), a reference video is the motion source (→ reference-motion-transfer), the user wants to edit an existing clip, or the user has exactly two images and nothing else (→ h3-fl2va-film, the FL2VA-only studio).

2. Intake

Confirm: ① the image(s) and their role (start frame / end frame / keyframe sequence), ② aspect ratio & duration (4–15s), ③ motion description with beats, ④ style & scene (comes from the image + style sentence), ⑤ on-screen text, ⑥ audio policy.

3. Goal

One approved video where the subject's identity matches the reference images in every frame and the motion follows the written beats exactly, ending per the mode contract (I2VA: first frame locked; FL2VA: both frames locked; L2VA: final frame locked).

4. Mode Selection

AssetsModeAlignment line
First-frame image + text motionI2VAFor the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.
Start + end pose imagesFL2VAHow the reference pictures align with the target video — Picture 1 (from Shot 1) aligns with the 0.00-second mark ...; Picture 2 (from Shot N) aligns with the S.SS-second mark ...
End-pose image onlyL2VAHow the reference pictures align with the target video — <Picture 1> (from [Shot N]) aligns with the S.SS-second mark ...
Up to 4 keyframes, sequentialI2VA multi-frameOne fully referenced clause per frame, each pinned to its timestamp

5. Prompt Construction

After the alignment line and a blank line, three fields:

  1. integrated_multimodal_description — style sentence → [Shot N] (no timestamp) → timecoded beats From X.XX to Y.YY seconds, .... Identity anchors (3–5 per subject: silhouette, colors, signature props) must recur across beats. Motion path structure: initial state → middle changes → gradual approach → final state landing on the end frame.
  2. overall_soundscape — per-beat diegetic sound.
  3. non_diegetic_music — genre, BPM, curve; N/A when unused.

Rules: end timestamp must equal the total duration with two decimals (S.SS); single-shot preference for FL2VA / L2VA; dialogue/on-screen text keep original language; 350–500 English words.

Ready-to-run examples: references/prompt-library.md.

6. Generation & Review

  • Model: MiniMax H3, ratio & duration as planned; 2K only after approval.
  • Checklist: identity holds in every frame (no feature leak from other reference images) / each beat appears in order with correct timing / end frame matches the locked picture / camera matches the written move / audio matches policy. Score pass / partial (name the beat) / fail.

7. Retry Strategy

SymptomFix
Identity drifts mid-videoAdd 1–2 identity anchors; re-state the image as a mid-video keyframe if I2VA multi-frame
End frame not landedRe-check the alignment line timestamp; extend the final approach beat
Motion too genericAdd timecoded micro-beats; shorten the windows
Wrong subject element leaks inRemove the leaked attribute from every beat; restate fully referenced clauses
Camera inconsistentState the move inside each beat sentence

Escalate after two failed retries of the same layer — offer fewer keyframes or a simpler motion path.

8. Output

Deliver: final video file; the prompt with alignment line exactly as used; settings record (mode, ratio, duration, audio policy); review notes.

Boundaries

Image-to-video only. No images → h3-promo-film. Reference-video motion → reference-motion-transfer. Do not animate real identifiable persons without consent, brand logos, or copyrighted characters; abstract or decline.

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexUnverifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall