SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsmedia-higgsfield-assets

media-higgsfield-assets

Agents & models

View repository

Updated 10-15 21:38

About this skill

Higgsfield assets and post-processing (media-higgsfield-assets)

moai-media | 3D · Audio · Video analysis · Post-processing (core: media-higgsfield-core)

Overview

media-higgsfield-image와 media-higgsfield-videoIf "Create new images and videos" handles that, this skill handles the four areas outside it: 3D assets, audio, finished video analysis, and post-processing of existing assets.

The call contract, cost preflight, and namespace resolution follow the core:

  • Call contract: ../media-higgsfield-core/references/call-schema.md
  • Live lookup: ../media-higgsfield-core/references/catalog-protocol.md
  • Jobs, costs, and readback: ../media-higgsfield-core/references/job-lifecycle.md

Trigger keywords

3D, GLB, mesh, rigging, skeleton, 3D modeling, texture, PBR, sound effects, SFX, ambience, background music, BGM, narration, TTS, speech synthesis, voice, viral prediction, hook score, video analysis, upscale, 4K, reframe, outpainting, background removal, cutout

Four areas and entry points

Area What it does Details
3D Image/text → GLB mesh, rigging, animation references/3d.md
Audio Sound effects, ambience, music, TTS references/audio.md
Analysis Attention, hook, and retention scores of finished videos references/analysis.md
Post-processing Upscaling, reframing, outpainting, background removal, motion §Post-processing below

Workflow

Follow the core REQ-010 flow exactly. Only the area differs; the order is the same.

  1. Intent → narrow down the area and candidates. Choose the area from the table above, and narrow down candidate models using the relevant reference file. Do not assume parameters.
  2. Live lookup. models_explore(action:'get')Fetch the actual constraints with . For 3D and audio, parameter variation between models is greater than for images and videos, so skipping this step almost always fails.
  3. Cost preflight. get_cost: true로 credits Confirm. For 3D, texture, rigging, and animation each cost extra, so check the cost with the options enabled.
  4. Approval gate. Stop before credits are spent — follow the core §Paid generation approval gate exactly. Show the full prompt, model, input media, options, count,adjustmentsand estimated credits, and obtain approval. If texture, rigging, and animation options were enabled for 3D, show the additional cost for each option separately — showing only the total makes it impossible to tell which option is expensive.
  5. Generation. Call only with approved values.
  6. Polling and readback. job_statusuntil completion. adjustmentsIf present, report it to the user.

Post-processing (modifying existing assets)

This is the case of modifying an existing asset rather than creating a new one. Each task has a dedicated tool, so do not try to reproduce the same result with a generative model — the dedicated tool is cheaper and the results are more stable.

Request Dedicated tool
Increase resolution (image) Image upscale
Increase resolution (video) Video upscale
Expand canvas / remove crop Outpainting
Change video aspect ratio (landscape↔portrait) Reframe
Remove background / transparent background Remove background
Motion transfer / recast / puppet Motion control

Input is passed in the same way as the core rules, via media_id or the previous job's job_id. Raw URLs are rejected.

Output format

## Higgsfield 에셋 생성 결과
- 영역: [3D | 오디오 | 분석 | 후처리]
- 모델·도구: [models_explore로 확인한 실제 id]
- 적용 옵션: [텍스처·리깅·애니메이션 / 포맷·샘플레이트 / 등]
- 비용: [get_cost가 반환한 credits]
- Job ID / 결과 URL: [job_status completed]
- 서버 조정(adjustments): [있으면 그대로 보고]

Cautions

  • 3D의 enable_animation은 enable_riggingis required, and texture-related options require should_texture. Violating the dependency relationships is an error (→ references/3d.md).
  • Rigging is tailored to humanoids. Animals and objects may produce poor rigging results, and this fact is disclosed in advance.
  • Some audio models are declared in the live schema as game pipeline only. Do not select them as the default for general-purpose requests (→ references/audio.md).
  • Video analysis does not create media and returns a text report. Do not let users expect generated output.
  • Do not guess model ids or parameters — always verify with models_explore.
  • Do not clone another person's voice without consent.

Related skills

Skill Perspective
moai-media:media-higgsfield-core Core: invocation contract, cost, namespace
moai-media:media-higgsfield-image Prerequisite: image generation for 3D input
moai-media:media-higgsfield-video Prerequisite: video generation for analysis and post-processing targets
moai-media:media-higgsfield-explainer Follow-up: assemble audio and clips into an explanatory video
moai-media:media-audio-gen Alternative: ElevenLabs-based TTS and dubbing
moai-marketer:marketing-performance-report Follow-up: reflect analysis scores in the performance report

Source

  • Higgsfield Skills (official agent documentation) — higgsfield-generate v0.12.0 (MIT). Basis for area classification and viral analysis interpretation criteria.
  • Live MCP schema observation (models_explore type=3d / type=audio) — basis for the model list, parameters, dependency relationships, and game-pipeline-only notation. Evidence tier: 1차.
  • The actual parameters are the runtime models_exploreis the only source of truth. Values in reference files are snapshots at the time of writing.