批量视频生成

概述

LoomLoom Batch Video Generator

Operate LoomLoom's official text-image-video-v1 template, whose official name is “通用文生图视频”. Treat “批量生成视频” as an alias for this exact template. Explain that the verified workflow generates an image first and then a short video; its outputs include both image and video artifacts.

Mandatory rules

• Never map the request to a similar template. Expected ID: text-image-video-v1; expected name: “通用文生图视频”.
• Never describe this as a pure text-to-video-only workflow. It is verified as text → image → video.
• Never invent fields, ratios, durations, models, audio behavior, fees, currencies, output paths, URLs, or run status.
• Never ask the user to paste a Token into chat or print, log, store, or repeat it.
• Never run env, printenv, set -x, or pass a real Token through visible --token arguments.
• Treat workbook download, editing, validation, and precheck as preparation. Treat submission as a potentially paid external action.
• Obtain fresh explicit confirmation immediately before submission. Invalidate it when any workbook, template, task-count, quote, or expiry detail changes.
• Use the live schema and downloaded official workbook as authority. Do not create a substitute or alter sheets, columns, option lists, or metadata.

Read [references/text-image-video-v1-contract.md](references/text-image-video-v1-contract.md) before data preparation. Read [references/cli-contract.md](references/cli-contract.md) before commands. Read [references/safety-and-billing.md](references/safety-and-billing.md) before submission.

Workflow

1. Introduce the capability

Explain briefly:

• “批量生成视频” maps to official text-image-video-v1 / 通用文生图视频.
• Each Excel row is one independent workflow that first creates an image, then turns it into a short video.
• Suitable scenarios include product concept clips, social-media short visuals, campaign materials, scene concepts, animated illustration drafts, storyboard motion samples, and batch creative exploration.
• The verified v3 template supports optional public reference-image URLs, horizontal/vertical video, 4/6/8-second duration, optional generated sound, and optional image/video model selection.

Ask what video scenes the user wants and approximately how many rows. Ask one missing question at a time.

2. Probe and authenticate

Resolve the Skill directory and run:

python3 scripts/probe_loomloom.py

If installed, run loomloom doctor. If missing, explain the verified stable installation route in [references/cli-contract.md](references/cli-contract.md) and obtain permission before changing the machine.

Prefer browser authentication:

loomloom login

Use loomloom login --no-browser if necessary. As a fallback, let the user configure LOOMLOOM_SERVER and LOOMLOOM_TOKEN privately outside chat. Verify only with loomloom doctor and report no credential values.

3. Verify the mapped template

Run only commands confirmed by local help:

loomloom template list -o json
loomloom template schema text-image-video-v1 -o json

Require ID text-image-video-v1, name “通用文生图视频”, and the nine verified v3 fields listed in [references/text-image-video-v1-contract.md](references/text-image-video-v1-contract.md). If absent or changed, stop and explain. Accept a newer version only after comparing its live fields, enum values, and workflow description.

4. Download and inspect the official workbook

loomloom template download text-image-video-v1

Capture the actual CLI path, preserve an untouched copy, and run:

python3 scripts/inspect_xlsx.py <downloaded.xlsx>

Confirm the nine data headers and workbook metadata ID before filling it.

5. Prepare video rows

Guide the user through the official columns:

• 画面描述 — required. Describe subject, scene, action, motion, camera movement, composition, lighting, color, mood, and desired ending when relevant.
• 视觉风格要求 — optional. Add photographic, cinematic, animation, rendering, color, texture, or art-direction preferences.
• 参考图片url — optional. Use one complete, publicly reachable http or https image URL only. Local paths, private URLs, multiple URLs, and data URLs are unsupported in verified v3.
• 图片比例 — required. Use 1:1, 4:5, 16:9, or 9:16.
• 图片生成模型（可选） — leave blank for the default or select only from the downloaded workbook dropdown. Keep one model consistent throughout the workbook.
• 视频比例 — required. Use 16:9 or 9:16.
• 视频时长（秒） — required. Use 4, 6, or 8.
• 是否生成声音 — required. Use false or true; do not promise narration, music, or a specific sound result.
• 视频生成模型（可选） — leave blank for the default or select from the workbook dropdown. Keep one model consistent throughout the workbook.

Preserve sheets, headers, validations, option lists, metadata, and formatting. Flag missing required fields, invalid enum values, inaccessible reference URLs, duplicate rows, conflicting model selections, unexpected formulas, and descriptions that omit the main subject or motion. Do not upload or submit during preparation.

6. Validate and precheck

Run local inspection, then authoritative server checks:

loomloom template validate-file text-image-video-v1 <xlsx-path>
loomloom template precheck-file text-image-video-v1 <xlsx-path>

Resolve all errors. Present exact template ID, absolute path, SHA-256, validated task count, selected duration/audio/model choices, quote total and currency, pricing basis, balance status, and quote expiry. This is a multi-stage media workflow; never estimate from image-only or video-only pricing.

If precheck is unavailable, state that no trustworthy estimate can be obtained. Do not submit unless the user explicitly accepts that uncertainty after seeing task count and settings.

7. Confirm the potentially paid run

After the summary, ask exactly:

回复“确认提交”后我才会开始执行本次 LoomLoom 批量视频任务。

Recompute SHA-256 immediately before submission. If anything changed, revalidate, requote, and reconfirm.

8. Submit and monitor

Derive a stable, non-secret idempotency key from text-image-video-v1 and the confirmed workbook SHA-256. Submit only after confirmation:

loomloom template submit-file text-image-video-v1 <xlsx-path> --client-request-id <stable-idempotency-key>

Capture the run ID. Do not retry an ambiguous submission with a new key; first inspect existing runs and reuse the original key only for an identical confirmed workbook.

Monitor:

loomloom run watch <run-id>

Report meaningful progress, completed and failed tasks, and terminal status. Do not call partial completion fully successful.

9. Download image and video results

Use locally supported commands:

loomloom run result-workbook <run-id>
loomloom run result-rows <run-id>
loomloom artifact list <run-id>
loomloom artifact download <run-id>

Report exact download paths, run ID, row-level errors, and which image/video files correspond to each input row. Explain that the result workbook preserves row alignment while artifacts contain generated media. Never invent filenames or URLs.

If some rows fail, prepare a workbook containing only failed rows and repeat validation, precheck, and confirmation. Never auto-rerun.

Failure handling

On unexplained failure, run loomloom doctor, sanitize output, and classify missing CLI, authentication, server, workbook, URL accessibility, balance, model, generation, or artifact issues. Preserve workbook, hash, idempotency key, and run ID. Stop at unsupported capabilities rather than improvising.