Skip to content
Back to search

VOX-COLLAGE-BROLL

Turn one spoken line into a paper-collage explainer video — Claude Code skill + MaxFusion MCP. English adaptation of gbro-collage-broll by pyang5166.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/megatroll222--vox-collage-broll/badge.svg)](/listings/megatroll222--vox-collage-broll)

HTML

<a href="/listings/megatroll222--vox-collage-broll"><img src="/api/listings/megatroll222--vox-collage-broll/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

1. Drop `SKILL.md` into your skills directory:

   mkdir -p ~/.claude/skills/collage-broll-explainers
   curl -o ~/.claude/skills/collage-broll-explainers/SKILL.md \
     https://raw.githubusercontent.com/MegaTroll222/VOX-COLLAGE-BROLL/main/SKILL.md

2. Connect the MaxFusion MCP server:

   claude mcp add --transport http maxfusion https://mcp.maxfusion.ai/mcp

The first call will ask you to authenticate with your MaxFusion account.

3. In Claude Code, say something like:

> *make a collage b-roll for: "Nobody misses washing dishes by hand."*

The skill proposes metaphors and takes it from there. When you want a voiceover, it lists the voices in your MaxFusion library and asks you to pick one.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 396 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: collage-broll-explainers
  3. description: Turn a ~5-second voiceover line, opinion sentence, or abstract concept into a premium editorial halftone paper-collage B-roll clip — optionally with a fitted voiceover. Load this skill whenever the user says "collage b-roll", "paper collage b-roll", "halftone collage", "collage-style visuals", "collage explainer" (or the Chinese triggers 纸拼贴 b-roll / 半调拼贴 / 拼贴风格配画面), or wants a script line turned into a collage visual metaphor. Enforces a mandatory three-gate approval flow — propose visual metaphors only, generate color collage stills only after the user confirms, and only after the stills are confirmed generate the assemble-from-empty animation. Video model is gemini-omni-flash, locked — no other video model is ever used and the choice is never offered to the user.
  4. ---
  5. # Collage B-roll Explainers
  6. Compress a ~5-second voiceover line into one sharp visual idea, then turn it into a premium editorial paper-collage assembly animation.
  7. Default pipeline:
  8. 1. Design the visual metaphor only, then wait for the user to confirm
  9. 2. Generate the final still frame only, then wait for the user to confirm
  10. 3. Automatically generate the video with first/last-frame references and run QA
  11. 4. Optionally generate a fitted voiceover and mux it in
  12. The two confirmation gates are part of the workflow, not overhead. They let the user spend their attention on aesthetics and direction, and prevent a wrong metaphor or a wrong still from burning video-generation cost.
  13. ## Runtime requirements
  14. - **MaxFusion MCP** connected and authenticated (`https://mcp.maxfusion.ai/mcp`). All image, video, and speech generation runs through it. If it is not connected, stop and walk the user through connecting before doing anything else.
  15. - **Local file access and ffmpeg** (e.g. Claude Code): frame preparation, video QA, silent delivery, and voiceover fitting download assets and process them locally. A sandboxed agent (no local downloads) can still run all three gates and display results in the MaxFusion widget, but must tell the user that frame preparation falls back to prompt-only color matching and that frame-by-frame self-QA, audio stripping, and voiceover fitting were skipped — the user should judge the clip by eye and mute it in their editor.
  16. ## Mandatory approval protocol
  17. ### Gate 1: Metaphor confirmation
  18. When you receive the script line(s), propose visual metaphors first. Do not generate any image, do not generate any video, do not call any generation tool.
  19. Deliver, for each line:
  20. - Core meaning
  21. - Emotion
  22. - One-sentence visual proposition
  23. - 3–6 key objects
  24. - Suggested background color and local accent colors
  25. - Intended assembly order
  26. Then stop explicitly and wait for the user to reply "approved", "all approved", or give per-item revision notes.
  27. If the user approves only some of the numbered items, only the approved items advance to Gate 2; keep iterating on the metaphors for the rest.
  28. ### Gate 2: Still-frame confirmation
  29. Only after the metaphor is confirmed do you write the visual spec and the image prompt, and generate the final still with `maxfusion_generate_image`.
  30. Save the originals into the project directory, build a numbered still contact sheet, display every still to the user (`maxfusion_display_generation`, one call per image, plus the numbered mapping), and stop again. At this stage you still do not generate any video.
  31. If the user approves only some stills, only the approved items advance to Gate 3; regenerate the rejected stills and re-confirm them first.
  32. ### Gate 3: Video generation
  33. Once the stills are confirmed, generate immediately. The video model is locked:
  34. ```text
  35. video_model: gemini-omni-flash
  36. ```
  37. No other video model is ever used. Do not offer a model choice, do not ask the user which model to use, do not fall back to anything else on failure — apply the repair playbook and re-run on gemini-omni-flash.
  38. ## Success criteria
  39. - One line expresses exactly one clear metaphor
  40. - A batch shares one design language, but is not forced onto a single blue background
  41. - The background is a strong, flat, uniform color field, varied by meaning
  42. - Subjects are built from black-and-white halftone photographic cut-outs
  43. - Key cards, buttons, film strips, rulebooks etc. may use red, yellow, teal, orange, purple, or cream colored cardstock
  44. - Every paper piece has crisp cut edges, a cream keyline, a soft low-opacity shadow, and paper grain
  45. - Motion is assemble-from-empty — not gentle drift, wobble, or slow zoom
  46. - No subtitles, no full script text, no logos, no watermark, no UI
  47. - Default delivery: 9:16, 5 seconds, 720p, silent MP4 (audio track stripped) — or the voiced version when the user wants voiceover
  48. ## When NOT to use this skill
  49. - Precise control of layers, occlusion, camera fly-throughs, or an editable timeline is needed: use a layered animation tool instead (e.g. an HTML-rendered layered video approach)
  50. - The user only wants a video prompt, not a finished clip: just write the prompt, skip this flow
  51. - The user needs a real-person product ad or an on-camera presenter: do not use this collage flow
  52. - The user explicitly needs per-layer editable transparent assets: this skill does not deliver separated transparent layers
  53. ## Default project directory
  54. Local working folder, named with the user's local date:
  55. ```text
  56. ~/collage-broll-projects/YYYY-MM-DD-<title>/
  57. ```
  58. Recommended structure for batch projects:
  59. ```text
  60. <project>/
  61. ├── brief.md
  62. ├── visual-spec.json
  63. ├── imagegen-prompts.md
  64. ├── omni-jobs.json
  65. ├── gate2-qa.md
  66. ├── gate3-qa.md
  67. ├── still-contact-sheet.jpg
  68. ├── omni-contact-sheet-all.jpg
  69. ├── video-first-frame-all.jpg
  70. ├── end-frame-comparison-all.jpg
  71. ├── 01-<concept-name>/
  72. │ ├── omni-prompt.txt
  73. │ ├── voiceover-raw.mp3 (optional voiceover stage)
  74. │ ├── voiceover-trim.mp3
  75. │ ├── frames/
  76. │ │ ├── last-frame-original.png
  77. │ │ ├── first-frame.png
  78. │ │ └── last-frame.png
  79. │ └── omni/run-v01/
  80. │ ├── final-5s.mp4
  81. │ ├── final-5s-noaudio.mp4
  82. │ ├── final-5s-voiced.mp4 (optional voiceover stage)
  83. │ ├── contact-sheet.jpg
  84. │ ├── video-last-frame.jpg
  85. │ └── end-frame-comparison.jpg
  86. └── 02-<concept-name>/...
  87. ```
  88. Also create a MaxFusion project via `maxfusion_create_project` (same name) and pass its id as `project_id` / `project_group_id` on every generation in the batch, so all assets stay grouped in the workspace. Track per item: the confirmed metaphor text, image job id + image id of the confirmed still, uploaded first/last frame `file_id`s, video job id + video id, and the QA verdict — and report these ids to the user with each gate so any item can be referenced or re-run individually.
  89. ## Phase 1: Design the visual metaphor
  90. First compress the script line into a single visual proposition.
  91. Extract:
  92. - Core meaning: what must the viewer finally understand
  93. - Emotion: calm, surprise, urgency, sudden clarity, absurdity, irony
  94. - Action verbs: open, connect, leak, bind, archive, light up, compress, fork, assemble
  95. - Visual metaphor candidates: machine, clock, film strip, filing cabinet, console, rulebook, funnel, rail track, chess piece
  96. Do not put the script text verbatim into the frame. Default to one metaphor per line, held to 3–6 key objects; too many elements dilutes the meaning and makes the Omni assembly unstable.
  97. Prefer metaphors a viewer recognizes from everyday life (a kettle, a snowball, a photo booth) over abstract ones — the spoken line does the bridging to the user's actual topic, so the visuals can stay wild without being obscure.
  98. For batches, prefer metaphors that form a before/after narrative — e.g. first manual toil and experience leaking away, then codified process and human–machine division of labor.
  99. ### Gate 1 output example
  100. ```text
  101. 1. Core meaning: expertise gets consumed from scratch every single time
  102. Visual metaphor: a skilled editor circles a giant film-strip clock cutting frame by frame; the clock completes a full revolution yet yields only a short strip of finished film
  103. Key objects: film-strip clock, editor figure, scissors, short output strip
  104. Colors: burnt-orange field, cream and light-teal accents
  105. Assembly order: clock → figure with scissors → film strip → final short output
  106. ```
  107. Output this and stop; wait for confirmation.
  108. ## Phase 2: Generate the color collage stills
  109. After the metaphor is confirmed, first write a self-contained `visual-spec.json`, then the image prompt.
  110. ### Visual spec
  111. ```json
  112. {
  113. "script_meaning": "",
  114. "visual_metaphor": "",
  115. "style_signature": "flat bold color field, mixed black-and-white halftone cut-outs and colored cardstock accents, crisp cut edges, cream keylines, soft paper shadows, editorial paper collage",
  116. "aspect_ratio": "9:16",
  117. "color_field": {
  118. "background_hex": "",
  119. "accent_colors": [],
  120. "paper_grain": "fine uncoated-paper fiber"
  121. },
  122. "elements": [
  123. {
  124. "what": "",
  125. "role": "",
  126. "motion": "",
  127. "placement": ""
  128. }
  129. ],
  130. "composition": {
  131. "layout": "",
  132. "negative_space": "",
  133. "final_frame": ""
  134. },
  135. "motion_plan": "structure first, subject or cards second, action and result last",
  136. "avoid": "typography, readable letters, numerals, logos, watermark, UI, subtitles, glossy 3D, photoreal environment"
  137. }
  138. ```
  139. ### Color rules
  140. Do not treat cobalt blue as the one default. Pick a strong color field from the meaning, and across a batch keep "same design language, different field colors":
  141. - Burnt orange / red: time consumption, labor, urgency
  142. - Mustard yellow: tools, warning, experience leaking away
  143. - Ink green: cognition, taste, system reset
  144. - Deep purple: rules, codification, long-term memory
  145. - Teal: judgment, collaboration, autonomous execution
  146. Subjects stay mostly black-and-white halftone, but every colored cardstock accent must serve the information hierarchy — never color for color's sake.
  147. ### Image prompt template
  148. Generate with `maxfusion_generate_image` (default model, `aspect_ratio: "9_16"`, `quality: "high"`):
  149. ```text
  150. Use case: ads-marketing
  151. Asset type: final still frame for a 9:16 image-to-video B-roll clip
  152. Primary request: Create a finished editorial paper-collage image expressing [one-sentence visual proposition].
  153. Scene/backdrop: perfectly flat [color name] paper field [hex] with subtle uncoated paper fiber.
  154. Style/medium: premium editorial stop-motion paper collage; black-and-white halftone photographic cut-outs mixed with selective [accent colors] colored cardstock.
  155. Composition/framing: vertical 9:16 locked poster frame; central subject within the middle 70 percent; generous clean color-field negative space; 3–6 large separable paper groups for later assemble-from-empty animation.
  156. Materials/textures: visible printed halftone dots, crisp machine-cut edges, thin warm-cream paper keylines, soft low-opacity physical drop shadows.
  157. Constraints: [the one relationship this metaphor must communicate at a glance].
  158. Avoid: no typography, no readable letters, no numerals, no logos, no watermark, no UI, no subtitles, no glossy 3D, no photoreal environment, no clutter.
  159. ```
  160. If the scene contains any dial, clock face, gauge, or counter, add an explicit constraint that every mark on it is a plain tick line — never Roman numerals or digits. If the scene contains a sleeping figure, explicitly ban "Zzz" marks. Video models love sneaking these in.
  161. Poll the job with `maxfusion_get_job` (`wait: 60`); download each still from its asset URL into `<item>/frames/last-frame-original.png` and record the image job id + image id.
  162. ### Still QA
  163. - Is the metaphor readable at a glance
  164. - Is the subject concentrated
  165. - Any fake lettering, logos, watermark, or UI
  166. - Enough pure color field left to assemble from an empty frame
  167. - 3–6 clear large groups, not a screen full of confetti
  168. - Across the batch: unified texture, varied field colors
  169. Copy the passing originals into the project directory, build a numbered still contact sheet, show it to the user (plus `maxfusion_display_generation` per image), and stop for Gate 2 confirmation. Write the still QA verdicts to `<project>/gate2-qa.md`.
  170. If the user asks to regenerate some stills, regenerate only those and produce `still-contact-sheet-v2.jpg` (v3, v4… on later rounds), keeping the old contact sheets for comparison — never overwrite them, and never reuse a rejected still downstream.
  171. ## Phase 3: Generate the video with Omni
  172. ### 1. Prepare first and last frames
  173. Keep the imagegen original, then normalize the last frame:
  174. ```bash
  175. ffmpeg -y -i <item>/frames/last-frame-original.png \
  176. -vf "scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920" \
  177. <item>/frames/last-frame.png
  178. ```
  179. The first frame is by default a pure empty paper field in the same background color as the last frame:
  180. ```bash
  181. ffmpeg -y -f lavfi -i color=c=0x<HEX>:s=1080x1920 \
  182. -frames:v 1 <item>/frames/first-frame.png
  183. ```
  184. (`<HEX>` is the `background_hex` from the visual spec — the same hex written into the image prompt, which is what keeps the two frames on the same color.)
  185. Only if the user explicitly asks not to start from a fully blank field may the first frame retain one base structural object.
  186. Upload both frames to MaxFusion with `maxfusion_upload_asset` (`purpose: "image_reference"`, `content_type: "image/png"`, exact `size_bytes`, `filename`): the tool returns an upload session — POST a multipart form to `upload.url` with every key/value from `upload.fields` exactly as given plus the file bytes as the final part named `file` (any 2xx = success), then use each returned `file_id`. Record the two `file_id`s per item.
  187. ### 2. Write the Omni animation prompt
  188. Default action order:
  189. ```text
  190. base structure → figure or key cards → connectors → action → final result
  191. ```
  192. Prompt template (save per item as `<item>/omni-prompt.txt`):
  193. ```text
  194. Paper-collage stop-motion assembly, using Image 1 as the exact empty first frame and Image 2 as the exact completed last frame. In one continuous locked-off vertical shot, open on the empty flat [color] paper field.
  195. Assemble the scene piece by piece with crisp physical stop-motion timing: [describe, in order, how the 3–6 elements slide in, snap into place, connect, and complete the action]. End by holding the supplied completed composition.
  196. Preserve the exact 9:16 framing, [hex] color field, colored cardstock accents, uncoated paper grain, halftone dots, cream keylines, crisp cut edges and soft shadows. Restrained tactile 2D paper craft only.
  197. No scene cuts, no camera movement, no zoom, no morphing, no new objects, no text, no letters, no numbers, no logos, no watermark, no UI, no sound.
  198. ```
  199. Every prompt must state that Image 1 is the empty first frame and Image 2 is the confirmed completed last frame. The final composition must land on Image 2 — do not let the model redesign it. Carry any anti-lettering constraint from the still prompt (tick-marks-only dials, no Zzz) into the video prompt too.
  200. ### 3. Submit the generation
  201. One `maxfusion_generate_video` call per item. The reference-image order carries the frame roles: the empty first frame is the FIRST id, the normalized last frame is the SECOND, matching "Image 1" / "Image 2" in the prompt.
  202. ```text
  203. video_model: gemini-omni-flash
  204. ref_image_file_ids: [<first-frame file_id>, <last-frame file_id>]
  205. prompt: <omni animation prompt>
  206. duration: 5
  207. aspect_ratio: 9_16
  208. resolution: 720p
  209. project_group_id: <batch project id>
  210. ```
  211. Record the submitted jobs in `<project>/omni-jobs.json` (prompt, frame file_ids, job id, output path per item). Poll each job with `maxfusion_get_job` (`wait: 60`) until terminal, then download each finished video to `<item>/omni/run-v01/final-5s.mp4`. If an individual job fails, re-run only that job — never re-run items that already passed.
  212. ### 4. Enforce silent delivery
  213. Even though the prompt says "no sound", still produce a zero-audio-track version with ffmpeg:
  214. ```bash
  215. ffmpeg -y -i <run>/final-5s.mp4 \
  216. -map 0:v:0 -c:v copy -an \
  217. <run>/final-5s-noaudio.mp4
  218. ```
  219. Deliver `final-5s-noaudio.mp4` by default; keep the original `final-5s.mp4` as an intermediate.
  220. ## Video QA
  221. Do not judge from the end frame alone — check the assembly progression and the final landing.
  222. ### Contact sheet
  223. ```bash
  224. ffmpeg -y -i <run>/final-5s-noaudio.mp4 \
  225. -vf "fps=1,scale=270:480,tile=5x1" \
  226. -frames:v 1 <run>/contact-sheet.jpg
  227. ```
  228. Pass criteria:
  229. - First frame is close to a pure empty color field; paper edges peeking in slightly at the borders are acceptable
  230. - Mid-clip shows structure, figure, or cards entering step by step — not a global fade-in
  231. - No cuts, no zoom, no 3D-ification, no drift into a photoreal scene
  232. - No fake lettering, logos, watermark, or UI
  233. - Final frame matches the confirmed still; minor pose or small-part drift (a slight figure pose change, a small part added or dropped) that does not affect the metaphor's meaning is a PASS — do not re-run for it
  234. - Output is 9:16, 720p, 5 seconds, zero audio streams (verify with ffprobe)
  235. Also extract the video's last frame and place it side by side with the confirmed still as `<run>/end-frame-comparison.jpg`. For batch projects, additionally merge three overview images:
  236. - `omni-contact-sheet-all.jpg`: per-second frames of every finished clip
  237. - `video-first-frame-all.jpg`: the actual first frame of every clip, verifying each truly opens on an empty color field
  238. - `end-frame-comparison-all.jpg`: confirmed stills side by side with video end frames
  239. Write the per-item QA verdicts — including the reasoning for any pass-with-noted-flaw — to `<project>/gate3-qa.md`.
  240. ### Repair playbook (common issues)
  241. - Paper edges visible in the first frame: slightly acceptable; if a strictly empty opening is required, prepend the opening beat in a video editor
  242. - Weak assembly feel: reduce the element count and rewrite the prompt as an explicit piece-by-piece slide-in / snap-into-place order
  243. - End-frame drift: strengthen "Image 2 is the exact completed last frame" and "End by holding the supplied completed composition"
  244. - Fake lettering appears (words, digits, Roman numerals on dials, "Zzz" on sleepers): go back and regenerate — a lettered still gets a new still; a lettered video on a clean still gets one video re-run with the explicit ban added to the prompt
  245. - One video fails: re-run only that job, on gemini-omni-flash, never by switching models
  246. ## Phase 4: Voiceover (optional)
  247. When the user wants the spoken line on the clip, generate it through MaxFusion and fit it locally.
  248. ### 1. Choose a voice
  249. List voices with `maxfusion_list_platform_voices` (and `maxfusion_list_my_voices` if the user has saved voices) and ask the user to pick one. Do not design a new voice unless the user explicitly asks. Reuse the same voice for every item in a batch.
  250. ### 2. Generate and trim
  251. Generate each line with `maxfusion_generate_speech` (the chosen `voice_id`, default `speed: 1.1` for a brisk read), download the mp3, and strip the trailing silence:
  252. ```bash
  253. ffmpeg -y -i <item>/voiceover-raw.mp3 \
  254. -af "areverse,silenceremove=start_periods=1:start_threshold=-45dB,areverse" \
  255. <item>/voiceover-trim.mp3
  256. ```
  257. ### 3. Fit the 4–5 second window
  258. Measure with ffprobe. The trimmed line must land between 4.0 and 5.0 seconds — shorter leaves dead air against the 5-second clip, longer gets cut off. TTS length varies ±1.5s between identical rolls, so always measure, never assume.
  259. Fix ladder, in order:
  260. 1. Trim silence (already done above)
  261. 2. Too short → lengthen the copy slightly; too long → tighten it; regenerate at `speed: 1`
  262. 3. Last resort for a slightly-long line: inaudible tempo nudge, `-af "atempo=1.05"` maximum
  263. ### 4. Mux
  264. Video stream is copied untouched; audio is padded to exactly 5 seconds:
  265. ```bash
  266. ffmpeg -y -i <run>/final-5s-noaudio.mp4 -i <item>/voiceover-trim.mp3 \
  267. -map 0:v:0 -map 1:a:0 -c:v copy -c:a aac -af apad -t 5 \
  268. <run>/final-5s-voiced.mp4
  269. ```
  270. ## Default delivery
  271. Deliver to the user:
  272. - Each item's `<item>/omni/run-v01/final-5s-noaudio.mp4` — or `final-5s-voiced.mp4` when the voiceover stage ran (and display the generation in the MaxFusion widget via `maxfusion_display_generation`)
  273. - Each item's contact sheet
  274. - The batch overview contact sheet
  275. - The end-frame comparison images
  276. - One sentence per item explaining how the script line became its visual metaphor
  277. - The batch's MaxFusion project name and the per-item ids needed to re-run or reference anything
  278. If a defect stems from Omni's fast-generation limits, say so plainly; only suggest switching to a layered animation tool when precise layer control is genuinely required.
  279. ## Credits
  280. The visual language, three-gate flow, prompt templates, color semantics, and QA/repair playbook originate from [gbro-collage-broll](https://github.com/pyang5166/gbro-collage-broll) by [pyang5166](https://github.com/pyang5166), originally written in Chinese for Codex with local Gemini scripts. This version is an English adaptation for Claude Code + MaxFusion MCP, with an added voiceover-fitting stage.
README.md · README · 79 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <pre>
  2. _____ _____ .__
  3. / \ _____ ___ ____/ ____\_ __ _____|__| ____ ____
  4. / \ / \\__ \ \ \/ /\ __\ | \/ ___/ |/ _ \ / \
  5. / Y \/ __ \_> < | | | | /\___ \| ( <_> ) | \
  6. \____|__ (____ /__/\_ \ |__| |____//____ >__|\____/|___| /
  7. \/ \/ \/ \/ \/
  8. </pre>
  9. # Collage B-roll Explainers
  10. A [Claude Code](https://claude.com/claude-code) skill that turns a single spoken sentence into a premium editorial **paper-collage explainer video** — 9:16, 5 seconds, halftone cut-outs assembling piece by piece on a bold color field, optionally with a fitted voiceover.
  11. You give it a line like *"A snowball only needs the first push — the hill does the rest."* It proposes a visual metaphor, waits for your approval, generates the finished still, waits again, then animates the whole scene assembling itself from an empty color field — and QAs its own output frame by frame.
  12. ## Examples
  13. | | | |
  14. |:---:|:---:|:---:|
  15. | ![snowball](examples/snowball-preview.gif) | ![candles](examples/candles-preview.gif) | ![dominoes](examples/dominoes-preview.gif) |
  16. | *"A snowball only needs the first push — the hill does the rest."* | *"One flame can light the whole room — and lose none of its fire."* | *"Tip the first tile — and the rest of your week runs itself."* |
  17. Full-quality MP4s (with voiceover) are in [`examples/`](examples/).
  18. ## What the skill enforces
  19. - **Three-gate approval flow** — metaphors first, stills second, video last. You approve at each gate; nothing burns generation cost on an unapproved idea.
  20. - **A locked visual language** — flat bold paper fields, black-and-white halftone photographic cut-outs, colored cardstock accents, crisp cut edges, cream keylines, paper grain. Color semantics are part of the system (burnt orange = urgency, deep purple = rules, teal = judgment…).
  21. - **Assemble-from-empty motion** — the clip opens on a pure empty color field and builds the scene piece by piece with stop-motion timing. First and last frames are prepared locally with ffmpeg and passed as tagged references, so the video lands exactly on the approved still.
  22. - **Self-QA** — contact sheets, first-frame purity checks, end-frame comparisons, an anti-fake-lettering rule set, and a repair playbook for the failure modes that actually happen.
  23. - **Voiceover fitting** (optional) — pick a voice from your library, generate the line, trim silence, fit it into a 4–5s window, and mux — video stream untouched.
  24. ## Requirements
  25. - **Claude Code** (or any agent with local file access) with **ffmpeg** installed — frame prep, QA, and muxing run locally on your machine.
  26. - **[MaxFusion](https://maxfusion.ai) MCP** — all image, video, and speech generation runs through it.
  27. ## Install
  28. 1. Drop `SKILL.md` into your skills directory:
  29. ```bash
  30. mkdir -p ~/.claude/skills/collage-broll-explainers
  31. curl -o ~/.claude/skills/collage-broll-explainers/SKILL.md \
  32. https://raw.githubusercontent.com/MegaTroll222/VOX-COLLAGE-BROLL/main/SKILL.md
  33. ```
  34. 2. Connect the MaxFusion MCP server:
  35. ```bash
  36. claude mcp add --transport http maxfusion https://mcp.maxfusion.ai/mcp
  37. ```
  38. The first call will ask you to authenticate with your MaxFusion account.
  39. 3. In Claude Code, say something like:
  40. > *make a collage b-roll for: "Nobody misses washing dishes by hand."*
  41. The skill proposes metaphors and takes it from there. When you want a voiceover, it lists the voices in your MaxFusion library and asks you to pick one.
  42. ## Using it without MaxFusion
  43. The skill file is written for the MaxFusion MCP tool surface, but the pipeline is portable. If you'd rather bring your own keys, you'll need:
  44. - a **Google AI Studio API key** — for the image generation (the stills) and the Gemini video model that does the assemble-from-empty animation from first/last-frame references, and
  45. - an **ElevenLabs API key** — for the voiceover stage.
  46. Swap the `maxfusion_*` tool calls in `SKILL.md` for your own scripts against those APIs; everything else (prompt templates, gates, ffmpeg frame prep, QA, the fix ladders) stays exactly the same. That is in fact how the original project worked — see credits below.
  47. ## Credits
  48. This skill is an English adaptation of **[gbro-collage-broll](https://github.com/pyang5166/gbro-collage-broll)** by **[pyang5166](https://github.com/pyang5166)**, who designed the visual language, the three-gate flow, the prompt templates, the color semantics, and the QA/repair playbook — originally in Chinese, built for Codex with local Gemini scripts.
  49. This version adapts it for Claude Code + MaxFusion MCP, translates it to English, and adds the voiceover-fitting stage. All credit for the core idea and the craft belongs to the original author.
  50. ## License
  51. [MIT](LICENSE)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.