luma-assets

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Luma Assets

Luma 资产管理

Use this skill when an agent needs to inspect, upload, name, classify, or reuse media assets for Luma workflows.

For local material libraries and PIP material matching, prefer

../luma-material/SKILL.md

当Agent需要为Luma工作流检查、上传、命名、分类或复用媒体资产时，可使用本Skill。

对于本地素材库和PIP素材匹配，优先使用

../luma-material/SKILL.md

。

Uploaded Video Intent

上传视频的意图确认

When the user uploads a video, do not guess its purpose. A video can be a voice sample, an avatar/source-role video, a PIP material, a reference for ASR/rewrite, or a video to process with subtitles/BGM/enhancement.

If the user did not clearly say the purpose, ask one short confirmation before running any destructive or paid workflow:

text

这段视频准备用来做什么？
1. 提取声音做音色
2. 当视频素材使用
3. 识别文案/仿写
4. 给视频加字幕/处理

Only proceed without asking when the wording is explicit:

Voice clone: "用这个声音", "克隆这个声音", "照这个人的声音", "用视频里的音频".
Avatar/source role: "用这个人出镜", "这个做数字人", "用这个视频当形象".
PIP/material: "当素材", "插到视频里", "素材库".
Reference/ASR/rewrite: "识别这段", "仿写这个", "提取文案".
Video processing: "加字幕", "加 BGM", "增强", "处理这个视频".

If the user wants voice clone from a video, treat the video as an audio source, not as a digital-human avatar. Extract or upload the audio sample for

voice.clone

; do not upload the video to

roles

unless the user explicitly asks to use the visual person as an avatar/source role.

当用户上传视频时，请勿猜测其用途。视频可以是语音样本、数字人形象/源角色视频、PIP素材、ASR/改写参考，或是需要添加字幕/背景音乐/增强处理的视频。

如果用户未明确说明用途，在执行任何具有破坏性或付费的工作流前，请先通过以下简短提问确认：

text

这段视频准备用来做什么？
1. 提取声音做音色
2. 当视频素材使用
3. 识别文案/仿写
4. 给视频加字幕/处理

只有当用户表述明确时，才可直接执行操作，无需询问：

语音克隆："用这个声音", "克隆这个声音", "照这个人的声音", "用视频里的音频"
数字人形象/源角色："用这个人出镜", "这个做数字人", "用这个视频当形象"
PIP/素材："当素材", "插到视频里", "素材库"
参考/ASR/改写："识别这段", "仿写这个", "提取文案"
视频处理："加字幕", "加 BGM", "增强", "处理这个视频"

如果用户希望通过视频进行语音克隆，请将视频视为音频源，而非数字人形象素材。提取或上传音频样本用于

voice.clone

；除非用户明确要求使用视频中的人物作为数字人形象/源角色，否则请勿将视频上传至

roles

分组。

Friendly Asset Names

友好的资产名称

Never expose hash-like object keys as the primary reusable name. Keep object keys internally, but give the user a short display name.

When a new video asset has a hash-like name or no useful filename:

Run asset/material understanding when available:

bash

luma-cli asset understand <object_name_or_key> --group <group> --output asset_meta.json

Generate a natural Chinese display name from the visual/audio summary, 5-10 Chinese characters.
Prefer concrete names that describe the person/scene/use, for example
```
老板在家里
```
,
```
女声装修讲解
```
,
```
门店口播素材
```
,
```
窗帘安装现场
```
.
Upload or save the asset with
```
--name <display_name>
```
when the command supports it.
Tell the user the display name and keep the object key only as technical detail.

If the generated name is uncertain, ask:

text

我先把它叫「老板在家里」，可以吗？

切勿将类似哈希值的对象键作为主要可复用名称对外展示。对象键仅在内部使用，需为用户提供简短的显示名称。

当新上传的视频资产名称为哈希值格式或无有效文件名时：

若支持资产/素材识别功能，执行以下命令：

bash

luma-cli asset understand <object_name_or_key> --group <group> --output asset_meta.json

根据视觉/音频摘要生成自然的中文显示名称，长度为5-10个汉字。
优先使用能描述人物/场景/用途的具体名称，例如
```
老板在家里
```
,
```
女声装修讲解
```
,
```
门店口播素材
```
,
```
窗帘安装现场
```
。
若命令支持，上传或保存资产时添加
```
--name <display_name>
```
参数。
将显示名称告知用户，仅将对象键作为技术细节留存。

若对生成的名称不确定，请询问用户：

text

我先把它叫「老板在家里」，可以吗？

Common Groups

常用分组

```
voice
```
: voice samples or generated voice assets
```
roles
```
: digital human avatar videos
```
asr_input
```
: uploaded media for transcription
```
enhance_input
```
: uploaded videos for enhancement
```
lipsync_input
```
: uploaded audio for lip-sync
```
default
```
: generic assets

```
voice
```
: 语音样本或生成的语音资产
```
roles
```
: 数字人形象视频
```
asr_input
```
: 用于转录的上传媒体
```
enhance_input
```
: 用于增强处理的上传视频
```
lipsync_input
```
: 用于唇形同步的上传音频
```
default
```
: 通用资产

Commands

命令示例

bash

luma-cli asset list voice
luma-cli asset list roles
luma-cli asset upload avatar.mp4 --group roles --name 老板在家里
luma-cli asset upload voice.wav --group voice --name 女声装修讲解
luma-cli asset delete upload8393 --group voice
luma-cli voice clone ./sample.wav --name my_voice
luma-cli material group list --output material_groups.json
luma-cli material group describe vlm_ai --output materials.json
luma-cli material search --materials materials.json --query "AI assistant" --limit 5 --output material_matches.json

bash

luma-cli asset list voice
luma-cli asset list roles
luma-cli asset upload avatar.mp4 --group roles --name 老板在家里
luma-cli asset upload voice.wav --group voice --name 女声装修讲解
luma-cli asset delete upload8393 --group voice
luma-cli voice clone ./sample.wav --name my_voice
luma-cli material group list --output material_groups.json
luma-cli material group describe vlm_ai --output materials.json
luma-cli material search --materials materials.json --query "AI assistant" --limit 5 --output material_matches.json

Agent Rules

Agent 规则

Prefer friendly names from
```
asset list
```
when available.
Prefer stable asset names or IDs returned by CLI commands.
For uploaded videos, confirm intent before choosing voice clone, avatar, material, ASR, or video-processing workflows.
Do not treat uploaded videos as digital-human avatar/source-role assets unless the user explicitly asks for the visual person to appear.
Do not upload a user-provided video with only a hash-like name. If no friendly name is available, generate one from understanding or ask the user to confirm a proposed name.
When uploading reusable assets, pass
```
--name <display_name>
```
and report that display name first.
Use a short display name for reusable assets; do not ask the user to remember hash-like object keys.
Ask for confirmation before deleting user assets, then use
```
luma-cli asset delete <name_or_stem> --group <group>
```
.
True in-place rename is not available yet. To change a name, re-upload or recreate the asset with
```
--name <display_name>
```
and delete the old one after user confirmation.
Upload local files before referencing them in cloud-only workflows.
Keep asset upload separate from creative workflow planning.
For local material libraries, prefer
```
material group describe
```
over hand-building a materials file.
Use
```
material search
```
before PIP planning when the script is long or the material group has many candidates.

优先使用
```
asset list
```
返回的友好名称。
优先使用CLI命令返回的稳定资产名称或ID。
对于上传的视频，在选择语音克隆、数字人形象、素材、ASR或视频处理工作流前，需先确认用户意图。
除非用户明确要求视频中的人物出镜，否则请勿将上传的视频视为数字人形象/源角色资产。
请勿仅以哈希值格式的名称上传用户提供的视频。若暂无友好名称，需通过识别生成或询问用户确认拟用名称。
上传可复用资产时，需传递
```
--name <display_name>
```
参数，并优先告知用户该显示名称。
为可复用资产使用简短的显示名称；请勿要求用户记忆类似哈希值的对象键。
删除用户资产前需先确认，然后使用
```
luma-cli asset delete <name_or_stem> --group <group>
```
命令。
目前暂不支持原地重命名。如需修改名称，需通过
```
--name <display_name>
```
重新上传或创建资产，并在用户确认后删除旧资产。
在仅支持云端的工作流中引用本地文件前，需先上传该文件。
将资产上传与创意工作流规划分开进行。
对于本地素材库，优先使用
```
material group describe
```
命令，而非手动构建素材文件。
当脚本较长或素材分组包含大量候选素材时，在规划PIP前需先使用
```
material search
```
命令。