agentleFS
frankxaifrankxai/creator-skills/skills/video/video-gen/SKILL.md

video-gen

Skill6 starsChanged 47 days agoMITView on GitHub

Route any image or video generation request to the right engine and pipeline. Use when asked to "make a video", "generate an image", "create b-roll", "make a video ad", "animate this photo", "clip for socials", "product shot", "talking head", "faceless video", "YouTube short", "UGC video", or any multi-step visual content request. Classifies intent by deliverable, identity consistency, and destination, dispatches to the engine skills you have installed (for example Higgsfield, Nano Banana, Grok Imagine), then chains assembly (HyperFrames) and editing/publishing (Descript). For pricing or capability ceilings, hand off to video-engine-routing.

  • Reads credentials

5 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the user did not specify duration, aspect ratio, or destination, infer from context.

    line 40
  2. Very specific:

    If the engine for a role is missing, use the next-best one you have (an image-only engine can cover stills, not clips).

    line 62

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

video-gen — the routing brain for visual content

One dispatcher, many engines. Intents stay stable; engines are swappable. When an engine changes, this file changes — the prompts and pipelines that call it do not.

Stack prerequisites

This skill routes to engines you install separately. It does not recommend a vendor. It uses the engines you already have, and when a role is missing it says which kind of engine unblocks you, then points to video-engine-routing for the cost and capability comparison.

EngineRoleInstall
Higgsfield CLI + skillsAll-in-one image and video engine (paid account); vendor skills: higgsfield-generate, higgsfield-soul-id, higgsfield-product-photoshoot, higgsfield-marketplace-cardsFollow the vendor's official install docs and read any install script before you run it
Gemini API keyNano Banana image gen and edits (character/reference work)GEMINI_API_KEY env var
Grok ImagineFast stylized images if you have a SuperGrok subscriptionvendor CLI/app
HyperFramesHTML video compositions, captions, TTS, audio-reactive motionnpx hyperframes + its skills
Descript MCPEdit by editing text: cuts, filler removal, captions, publishconnect the Descript MCP server

Any other image or video engine that fits a role below (a pay-per-use API such as fal.ai, a Veo or Kling account, cloud or local ComfyUI) works the same way: assign it to the role it covers.

Step 1 — classify the request

Three axes decide everything downstream:

AxisValues
Deliverablestill image · single clip (≤15s) · assembled video (edited, multi-scene)
Identity consistencynone · product (same object every shot) · face (same person every shot)
Destinationsocial feed · marketplace listing · YouTube · website/hero · ad campaign

If the user did not specify duration, aspect ratio, or destination, infer from context; ask only when the answer changes the engine choice.

Step 2 — dispatch

Pick the role first, then use the engine you have for it. The last column lists the shortcut when that vendor's skill is installed.

IntentRoleWhyShortcut if installed
Product/brand stillimage engine with product-shot templatesstudio, lifestyle, hero, carousel promptshiggsfield-product-photoshoot
Marketplace listing setimage engine with listing templatescompliant main image + secondary + A+ moduleshiggsfield-marketplace-cards
Same face across shotsidentity engine (train a reference once, reuse it)identity persists across images and videohiggsfield-soul-id → higgsfield-generate with the Soul id; a LoRA on ComfyUI does the same job
Generic still, design, legible textimage engine with strong text renderinglegible text is the deciding factorhiggsfield-generate (GPT Image 2)
Character/reference image editreference-following image editorbest at keeping the reference intactNano Banana via Gemini API
Fast stylized stillfast image enginezero marginal cost if already subscribedGrok Imagine (included in a SuperGrok subscription)
Single video clip, defaultbudget video enginebest price/quality for social clipshiggsfield-generate (Seedance 2.0)
Single video clip, cinematiccinematic video enginemotion quality, camera controlhiggsfield-generate (Kling 3.0)
Branded ad with avatar/productad-format video enginehooks, avatars, product placement in one flowHiggsfield Marketing Studio (via higgsfield-generate)
Captions, title cards, audio-reactive motion, scroll videoHyperFrames compositiondeterministic HTML video, no generation costHyperFrames
Cut, rearrange, remove filler, subtitle, publishtext-based editortext-based editing beats timeline scrubbingDescript MCP (import_media → prompt_project_agent → export)

If the engine for a role is missing, use the next-best one you have (an image-only engine can cover stills, not clips). If no video engine is installed, stop, say which kind of engine unblocks the request, and hand off to video-engine-routing.

Step 3 — pipelines

Steps name the role; the engine in brackets is the shortcut when installed.

30-second product ad

  1. Image engine, product-shot templates → 3 stills (hero, lifestyle, closeup) [higgsfield-product-photoshoot]
  2. Video engine, image-to-video on the hero, 5s [higgsfield-generate with Seedance 2.0]
  3. HyperFrames: title card + price overlay + end card around the clip
  4. Descript: assemble, caption, export per destination aspect ratio

Faceless YouTube b-roll batch

  1. Script beats → one prompt per beat
  2. Video engine, batch, 5s each [higgsfield-generate with Seedance 2.0]
  3. Voiceover: HyperFrames TTS, or ElevenLabs if the user has it
  4. Descript: import all, sequence to narration, auto-captions

Talking-head with a consistent face

  1. Identity engine: train once, store the reference id [higgsfield-soul-id]
  2. Video engine with that identity → presenter clips [higgsfield-generate with the Soul id]
  3. Descript: filler-word removal, captions, publish

Caption-heavy short (quote/hook format)

  1. Still from the dispatch table (or user-supplied)
  2. HyperFrames: audio-reactive captions, marker highlights, beat-synced motion
  3. Render via HyperFrames; no Descript needed

Quality gate before publishing

For anything going to a paid placement or a growth channel, check the finished cut for hook strength, retention risk and distraction. If your engine has a predictor, use it (Higgsfield's is brain_activity); otherwise review the first 2 seconds by hand against the same three points. Score low → fix the first 2 seconds before touching anything else.

When you hit a ceiling

Credits exhausted, need a custom pipeline (LoRA, ControlNet, motion transfer), or per-clip cost matters at volume → hand off to the video-engine-routing skill. It holds the decision tree and the cost structure for subscription vs pay-per-use vs cloud ComfyUI.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.