Ready to supercharge your AI agents?

Install any skill in seconds. Free to start, no credit card required.

Neodomain AI Content Generator · leoyeai | DecimalAI

leoyeai/neodomain-ai

SkillScore92Maintained

Generate images and videos via Neodomain AI API. Supports text-to-image, image-to-video, text-to-video, universal multi-modal video, motion control video, and batch storyboard video generation. Use when user wants to create AI-generated images or videos using the Neodomain platform.

design-mediaby leoyeai

Links this skill to your agents and keeps it current with the author’s updates.

Automatic

AI Engineer pack

1 · One-time — install the CLI (docs ↗)

pip install decimalai

2 · Get the skill — writes .claude/skills/leoyeai-neodomain-ai/SKILL.md

decimalai skills pull leoyeai-neodomain-ai --out .claude/skills/

No CLI? Get the same file — curl · raw file

curl -fsSL --create-dirs -o .claude/skills/leoyeai-neodomain-ai/SKILL.md https://app.decimal.ai/s/leoyeai-neodomain-ai/SKILL.md

Raw file: /s/leoyeai-neodomain-ai/SKILL.md ↗ · machine-readable summary at /s/leoyeai-neodomain-ai.json ↗

In productionGlobal

how this skill behaves on real traffic · last 30 days

Runs

427

all versions · 30d

Eval pass

85%

global

Used when offered

100%

used / offered · all versions

By model· real traffic, every org · 30d — not the benchmark

Model	Eval pass	Runs
gemini-3.6-flash	100%	43

Anonymized aggregate across every org using this skill — proof from real production use, not a demo.

Does this skill help? · Benchmark: with vs. withoutsign in to run this on your agents

23 test cases from the publisher’s eval.yaml · each run twice on gemini-3.6-flash — with the skill injected, then without · last run 50d ago

Yes — +83% lift · at higher cost+393% tokens0% turns96% of whole cases pass with skill (22/23)

Test case	Without → With	Effect	Δ tokens	Δ turns
case-01	✗→✓	▲ Improved	457%	0%
case-02	✗→✓	▲ Improved	381%	0%
case-03	✗→✓	▲ Improved	229%	0%
case-07	✗→✓	▲ Improved	446%	0%
case-08	✗→✓	▲ Improved	231%	0%

Showing 5 of 23 cases

SkillSafety

✓ 3 of 3 passed

Code scan15 of 15 checks clean

0 critical · 0 warning · 0 info — scanner v4 — 8/2/2026

live_secretremote_code_executionreverse_shelldata_exfiltrationssrf_metadataagent_snoopinganti_refusaldestructive_commandsobfuscated_unicodeunicode_homoglyphopaque_blobsuspicious_urltool_overreachtrigger_abuseprompt_injection_phrasing

IntentSafe

LLM judge · does the skill do what it claims

ContentClean

LLM judge · policy & harmful-content review

Source

Imported

LeoYeAI / openclaw-master-skills

skills/neo-ai/SKILL.md

View on GitHub ↗

License MITImported 8/2/2026Version v1Bundled 10 files · scripts

Skill contents

scripts/

SKILL.md

Neodomain AI Content Generator

核心行为规范

Token 检查：每次执行前确认 NEODOMAIN_ACCESS_TOKEN 已设置。若未设置或遇到 token 过期错误（errCode 2001），立即走认证流程，动态询问用户手机号/邮箱，不要硬编码保存。
模型选择：根据用户需求按模型选择逻辑自主选择，无需询问用户（除非用户明确指定）。
提示词规范：必须遵循各模式的提示词公式，尤其是 UNIVERSAL_TO_VIDEO 必须用 @图片 1/@视频 1 等明确引用素材。
输出目录：默认使用 ./output，若用户未指定则不需询问。
--generate-audio 是布尔 flag，写 --generate-audio 即可，不要写 --generate-audio true。

⚠️ 常见问题与解决方案 (FAQ)

1. Token has been revoked (errCode 2001)

现象：视频生成 API 返回 Token has been revoked，但图片 API 可能正常。

原因：

企业版 token 的视频服务权限被撤销
Token 已过期
服务端认证系统临时故障

解决方案：

先用图片 API 测试 token 是否有效：curl -H "accessToken: $TOKEN" https://story.neodomain.cn/agent/ai-image-generation/models
如果图片 API 正常但视频 API 失败 → 联系管理员确认视频服务权限
如果全部失败 → 重新登录获取新 token（走认证流程）

2. 模型不支持某 generationType

现象：❌ Failed: 模型 xxx 不支持 YYY 类型

原因：不是所有模型都支持所有生成类型。

解决方案：

neo-video-2-0 / neo-video-2-0-fast → 仅支持 UNIVERSAL_TO_VIDEO
kling-video-o1 / kling-v3-omni → 仅支持 UNIVERSAL_TO_VIDEO
veo-3.1-* → 支持 REFERENCE_TO_VIDEO（多图参考）
其他模型 → 通常支持 TEXT_TO_VIDEO 和 IMAGE_TO_VIDEO

正确选择：

多张角色图 → 用 neo-video-2-0 + UNIVERSAL_TO_VIDEO
单张首帧 → 用 doubao-seedance-1-5-pro + IMAGE_TO_VIDEO
纯文字 → 用 doubao-seedance-1-5-pro + TEXT_TO_VIDEO

3. UNIVERSAL_TO_VIDEO 提示词写法错误

错误示例：--prompt "蓝白校服的男生和穿碎花裙子的女生在校园里散步"

正确示例：--prompt "参考@图片 1 的男生和@图片 2 的女生，两人在阳光明媚的校园里并肩散步，镜头缓慢跟随"

规则：

--image-urls 中第 1 个 URL → 用 @图片 1 引用
--image-urls 中第 2 个 URL → 用 @图片 2 引用
--video-urls 中第 1 个 URL → 用 @视频 1 引用
--audio-urls 中第 1 个 URL → 用 @音频 1 引用
必须在 prompt 中明确引用每个素材，不要用"那张图"等模糊表达

4. 认证服务不可用

现象：Authentication service unavailable, please try again

原因：Neodomain 服务端认证系统临时故障（可能在重启）

解决方案：等待 5-10 分钟后重试

意图识别与路径选择

根据用户意图选择正确路径：

| 用户意图 | 路径 | |---------|------| | 生成图片（无参考图） | generate_image.py → TEXT_TO_IMAGE | | 生成图片（有参考图/角色一致性） | generate_image_ref.py | | 生成视频（纯文字描述） | generate_video.py → TEXT_TO_VIDEO | | 生成视频（有参考图/首帧） | generate_video.py → IMAGE_TO_VIDEO | | 生成视频（多张参考图保持一致性） | generate_video.py → REFERENCE_TO_VIDEO（仅 Veo 3.1） | | 生成视频（多模态：图 + 视频 + 音频组合） | generate_video.py → UNIVERSAL_TO_VIDEO（Universal 模型） | | 动作迁移（把参考视频动作转移到图片人物上） | motion_control.py | | 故事板/分镜批量生视频 | batch_video.py |

Model Selection Logic

图片模型选择

有参考图？
├─ 是 → generate_image_ref.py，模型用 doubao-seedream-5-0-260128
└─ 否 → generate_image.py
        ├─ 追求高性价比 → doubao-seedream-5-0-260128 (25pts)
        ├─ 追求最高质量 → gemini-3-pro-image-preview (100pts)
        └─ 默认/普通需求 → gemini-3.1-flash-image-preview (90pts)

视频模型选择

有多模态素材（视频/音频参考）？
└─ 是 → UNIVERSAL_TO_VIDEO
        ├─ 追求最高品质 → neo-video-2-0
        ├─ 追求速度/性价比 → neo-video-2-0-fast
        └─ 需要高分辨率 (1080p) → kling-v3-omni 或 kling-video-o1

有多张参考图（保持角色一致性）？
└─ 是 → REFERENCE_TO_VIDEO，模型用 veo-3.1-generate-preview

有参考图（图生视频）？
└─ 是 → IMAGE_TO_VIDEO
        ├─ 需要音频 → doubao-seedance-1-5-pro-251215 或 kling-v3
        ├─ 需要首尾帧精确控制 → kling-v3 或 kling-v2-6
        ├─ 需要超高清 4K → veo-3.1-generate-preview
        ├─ 需要超长时长 (15s+) → kling-v3(最长 15s) 或 vidu-q3-pro(最长 16s)
        └─ 默认 → doubao-seedance-1-5-pro-251215

纯文字生视频？
└─ 是 → TEXT_TO_VIDEO
        ├─ 需要音频 → doubao-seedance-1-5-pro-251215 或 kling-v3
        ├─ 高性价比 → doubao-seedance-1-0-pro-fast-251015
        └─ 默认 → doubao-seedance-1-5-pro-251215

视频参数选择规则

resolution：用户未指定时默认 720p；明确要求高清用 1080p；明确要求超清用 4K（仅 Veo 3.1 支持）
duration：用户未指定时默认 8s；需注意各模型上限（见下表），不要超出范围
aspect-ratio：根据用途自主判断——竖屏内容用 9:16，横屏用 16:9，方形用 1:1；IMAGE_TO_VIDEO 模式下此参数无效，无需传入
generate-audio：用户提到"有声音"/"带音频"/"有背景音乐"时自动加上

Authentication

bash
# Step 1：发送验证码（询问用户手机号或邮箱后执行）
python3 {baseDir}/scripts/login.py --send-code --contact "手机号或邮箱"

# Step 2：用户提供验证码后登录
python3 {baseDir}/scripts/login.py --login --contact "手机号或邮箱" --code "验证码"

Step 2 可能出现两种结果：

单一身份：直接输出 accessToken，告知用户设置到 NEODOMAIN_ACCESS_TOKEN 环境变量。

多身份：输出身份列表（含编号、类型、昵称、企业名、userId），需执行 Step 3 选择身份：

bash
# Step 3：选择身份（仅在 Step 2 返回多身份时执行）
python3 {baseDir}/scripts/login.py --select-identity --contact "手机号或邮箱" --user-id "选中的userId"

Step 3 成功后输出 accessToken，告知用户设置到 NEODOMAIN_ACCESS_TOKEN 环境变量。

Image Generation

generate_image.py（文生图 / 图生图）

bash
python3 {baseDir}/scripts/generate_image.py \
  --prompt "A beautiful mountain landscape, golden hour, cinematic" \
  --model "doubao-seedream-5-0-260128" \
  --aspect-ratio "16:9" \
  --size "2K" \
  --token $NEODOMAIN_ACCESS_TOKEN

| 参数 | 说明 | 默认值 | |------|------|--------| | --prompt | 图片描述 | 必填 | | --model | 模型 ID | gemini-3.1-flash-image-preview | | --aspect-ratio | 1:1, 16:9, 9:16, 4:3, 3:4 | 1:1 | | --num-images | 1 或 4 | 1 | | --size | 1K, 2K, 4K | 2K | | --negative-prompt | 负向提示词 | - | | --guidance-scale | 提示词遵循度 (1.0-20.0) | 7.5 | | --seed | 随机种子 | 随机 | | --output-format | jpeg, png, webp | jpeg | | --output-dir | 保存目录 | ./output |

可用模型：

| 模型 ID | 费用 | 支持尺寸 | 特点 | |--------|------|----------|------| | gemini-2.5-flash-image | 30 pts | 1K | 快速低价 | | gemini-3.1-flash-image-preview | 90 pts | 1K, 2K, 4K | 均衡 | | gemini-3-pro-image-preview | 100 pts | 2K, 4K | 最高质量 | | doubao-seedream-5-0-260128 | 25 pts | 2K, 3K | 最高性价比，推荐参考图用 | | doubao-seedream-4-5-251128 | 30 pts | 2K, 4K | - | | doubao-seedream-4-0-250828 | 25 pts | 2K, 4K | - | | gpt-image-1.5 | 100 pts | 1K | GPT 风格 | | Midjourney-niji 7 | 30 pts | 1K | 二次元/动漫 | | Midjourney-v 7 | 30 pts | 1K | 艺术风格 |

generate_image_ref.py（参考图生图 / 角色一致性）

bash
python3 {baseDir}/scripts/generate_image_ref.py \
  --prompt "A woman walking in a forest" \
  --reference-image "https://example.com/character1.jpg" \
  --reference-image "https://example.com/character2.jpg" \
  --model "doubao-seedream-5-0-260128" \
  --aspect-ratio "16:9" \
  --token $NEODOMAIN_ACCESS_TOKEN

--reference-image 可多次指定，上限 10 张。其余参数同 generate_image.py，默认模型为 doubao-seedream-5-0-260128。

Video Generation

generate_video.py

统一入口，通过 --generation-type 区分模式。

TEXT_TO_VIDEO（文生视频）

bash
python3 {baseDir}/scripts/generate_video.py \
  --prompt "A serene lake at dawn, mist rising, cinematic wide shot" \
  --model "doubao-seedance-1-5-pro-251215" \
  --generation-type "TEXT_TO_VIDEO" \
  --aspect-ratio "16:9" \
  --resolution "720p" \
  --duration "8s" \
  --generate-audio \
  --token $NEODOMAIN_ACCESS_TOKEN

IMAGE_TO_VIDEO（图生视频）

bash
# 首帧必填，尾帧可选（需模型支持首尾帧）；宽高比自动按参考图，无需指定
python3 {baseDir}/scripts/generate_video.py \
  --prompt "The camera slowly pans across the landscape" \
  --model "kling-v3" \
  --generation-type "IMAGE_TO_VIDEO" \
  --first-frame "https://example.com/image.jpg" \
  --last-frame "https://example.com/end.jpg" \
  --resolution "1080p" \
  --duration "5s" \
  --generate-audio \
  --token $NEODOMAIN_ACCESS_TOKEN

REFERENCE_TO_VIDEO（多图参考生视频）

仅 veo-3.1-generate-preview / veo-3.1-fast-generate-preview 支持。

bash
python3 {baseDir}/scripts/generate_video.py \
  --prompt "Characters interacting in a scene" \
  --model "veo-3.1-generate-preview" \
  --generation-type "REFERENCE_TO_VIDEO" \
  --image-urls "https://example.com/ref1.jpg,https://example.com/ref2.jpg" \
  --aspect-ratio "16:9" \
  --resolution "1080p" \
  --duration "8s" \
  --generate-audio \
  --token $NEODOMAIN_ACCESS_TOKEN

UNIVERSAL_TO_VIDEO（多模态组合生视频）

支持图片、视频、音频混合参考。必须在 prompt 中用 @图片 1/@图片 2/@视频 1/@音频 1 精确引用对应素材（顺序即编号）。

bash
python3 {baseDir}/scripts/generate_video.py \
  --prompt "参考@图片 1 的男生和@图片 2 的女生，两人在阳光明媚的校园里并肩散步，镜头缓慢跟随" \
  --model "neo-video-2-0" \
  --generation-type "UNIVERSAL_TO_VIDEO" \
  --image-urls "https://example.com/boy.jpg,https://example.com/girl.jpg" \
  --resolution "720p" \
  --duration "5s" \
  --token $NEODOMAIN_ACCESS_TOKEN

Universal 提示词公式： @素材引用 + 主体描述 + 动作/场景 + 镜头语言 + 风格美学

@素材引用规则（重要！）：

--image-urls 中第 1 个 URL → 用 @图片 1 引用
--image-urls 中第 2 个 URL → 用 @图片 2 引用
--video-urls 中第 1 个 URL → 用 @视频 1 引用
--audio-urls 中第 1 个 URL → 用 @音频 1 引用
必须在 prompt 开头明确引用每个素材，不要用"那张图""这个角色"等模糊表达

错误示例 ❌：

bash
--prompt "蓝白校服的男生和穿碎花裙子的女生在校园里散步"
--image-urls "boy.jpg,girl.jpg"

正确示例 ✅：

bash
--prompt "参考@图片 1 的蓝白校服男生和@图片 2 的碎花裙子女生，两人在阳光明媚的校园里并肩散步，镜头缓慢跟随"
--image-urls "boy.jpg,girl.jpg"

适用模型：neo-video-2-0, neo-video-2-0-fast, kling-video-o1, kling-v3-omni

generate_video.py 完整参数表

| 参数 | 说明 | 默认值 | |------|------|--------| | --prompt | 视频描述 | 必填 | | --model | 模型 ID | doubao-seedance-1-5-pro-251215 | | --generation-type | TEXT_TO_VIDEO / IMAGE_TO_VIDEO / REFERENCE_TO_VIDEO / UNIVERSAL_TO_VIDEO | TEXT_TO_VIDEO | | --first-frame | 首帧图片 URL（I2V 必填） | - | | --last-frame | 尾帧图片 URL（可选，需模型支持） | - | | --image-urls | 参考图 URL 逗号分隔（R2V 必填；U2V 可选，→ imageUrls） | - | | --video-urls | 参考视频 URL 逗号分隔（U2V 可选，→ referenceVideoUrls） | - | | --audio-urls | 参考音频 URL 逗号分隔（U2V 可选，→ audioUrl） | - | | --aspect-ratio | 16:9/9:16/1:1/4:3/3:4/21:9/9:21（T2V/R2V 有效；I2V 无效） | 16:9 | | --resolution | 480p/720p/768p/1080p/2k/4K（各模型范围不同） | 720p | | --duration | 时长如 5s/8s/10s（各模型范围不同） | 8s | | --fps | 帧率 | 24 | | --seed | 随机种子 | 随机 | | --generate-audio | 生成音频（flag） | false | | --enhance-prompt | 增强提示词（flag） | false | | --negative-prompt | 负向提示词 | - | | --output-dir | 保存目录 | ./output |

各模型支持范围（Cascading）

| 模型 ID | 类型 | 分辨率 | 时长 | 音频 | 首尾帧 | |--------|------|--------|------|------|--------| | doubao-seedance-1-5-pro-251215 | T2V/I2V | 480p, 720p, 1080p | 4–12s | ✅ | ✅ | | doubao-seedance-1-0-pro-250528 | T2V/I2V | 720p, 1080p | 2–10s | ❌ | ✅ | | doubao-seedance-1-0-pro-fast-251015 | T2V/I2V | 720p, 1080p | 3–10s | ❌ | ❌ | | kling-v3 | T2V/I2V | 720p, 1080p | 3–15s | ✅ | ✅ | | kling-v2-6 | T2V/I2V | 1080p | 5s, 10s | ✅ | ✅ | | kling-v2-5-turbo | T2V/I2V | 720p, 1080p | 5s, 10s | ❌ | ✅ | | kling-v2-1 | I2V 仅 | 720p, 1080p | 5s, 10s | ❌ | ✅ | | vidu-q3-pro | T2V/I2V | 720p, 1080p, 2k¹ | 5–16s | ✅ | ✅ | | vidu-q2-pro | I2V 仅 | 720p, 1080p | 5s, 8s | ❌ | ✅ | | wan2.6 | T2V/I2V | 720p, 1080p | 5s, 10s, 15s | ✅ | ❌ | | wan2.6-i2v-flash | I2V 仅 | 720p, 1080p | 5s, 10s, 15s | ✅ | ❌ | | MiniMax-Hailuo-2.3 | T2V/I2V | 768p, 1080p | 6s, 10s | ❌ | ❌ | | MiniMax-Hailuo-2.3-Fast | T2V/I2V | 768p, 1080p | 6s, 10s | ❌ | ❌ | | MiniMax-Hailuo-02 | T2V/I2V | 768p, 1080p | 6s, 10s | ❌ | ✅ | | veo-3.1-generate-preview | T2V/I2V/R2V | 720p, 1080p, 4K | 4s, 6s, 8s | ✅ | ✅ | | veo-3.1-fast-generate-preview | T2V/I2V/R2V | 720p, 1080p, 4K | 4s, 6s, 8s | ✅ | ✅ |

> ¹ Vidu Q3 Pro 的 2k 仅在 I2V 模式下支持。 > T2V 宽高比：Doubao 支持 16:9/9:16/1:1/4:3/3:4/21:9/9:21；Kling/Vidu/Wan 支持 16:9/9:16/1:1；Hailuo 仅 16:9；Veo 支持 16:9/9:16。

各模型支持范围（Universal，--generation-type UNIVERSAL_TO_VIDEO）

| 模型 ID | 分辨率 | 时长 | 音频 | 首尾帧 | 最大参考数 | |--------|--------|------|------|--------|-----------| | neo-video-2-0 | 480p, 720p | 4–15s | ✅ | ✅ | 图×9, 视频×3, 音频×3 | | neo-video-2-0-fast | 480p, 720p | 4–12s | ✅ | ✅ | 图×9, 视频×3, 音频×3 | | kling-video-o1 | 720p, 1080p | 3–10s | ❌ | ❌ | 图×12, 视频×1 | | kling-v3-omni | 720p, 1080p | 3–15s | ✅ | ❌ | 图×12, 视频×1 |

motion_control.py（动作迁移）

将参考视频的人物动作迁移到参考图片上：

bash
python3 {baseDir}/scripts/motion_control.py \
  --image "https://example.com/ref_image.jpg" \
  --video "https://example.com/ref_video.mp4" \
  --prompt "Make the character dance" \
  --mode "pro" \
  --duration 5000 \
  --token $NEODOMAIN_ACCESS_TOKEN

| 参数 | 说明 | 默认值 | |------|------|--------| | --image | 参考图片 URL | 必填 | | --video | 参考动作视频 URL | 必填 | | --prompt | 附加文字控制 | - | | --mode | std（标准）/ pro（专业） | std | | --orientation | 角色朝向来源：image / video | image | | --keep-original-sound | 保留原视频声音：yes / no | yes | | --duration | 时长（毫秒） | 5000 | | --output-dir | 保存目录 | ./output |

batch_video.py（批量分镜生视频）

> 需要 pip install oss2（用于上传本地图片到 OSS）。默认使用 doubao-seedance-1-5-pro-251215，IMAGE_TO_VIDEO，720p/5s。

bash
python3 {baseDir}/scripts/batch_video.py \
  --storyboard-dir ./storyboards \
  --output-dir ./output/videos \
  --start 1 \
  --end 17 \
  --token $NEODOMAIN_ACCESS_TOKEN

| 参数 | 说明 | 默认值 | |------|------|--------| | --storyboard-dir | 分镜图片目录（命名格式：shot_01_*.jpeg） | 必填 | | --output-dir | 视频输出目录 | 必填 | | --start | 起始镜头编号 | 1 | | --end | 结束镜头编号 | 17 |

Output

所有脚本完成后自动下载到输出目录：

图片：image_1.jpg / .png / .webp
视频：video.mp4 + thumbnail.jpg
批量视频：video_01.mp4, video_02.mp4, ...
metadata.json：任务 ID、参数、文件 URL 等详情

Illustrative Want to see what leoyeai/neodomain-ai would put in front of your model? Preview the rendered body and check its trigger phrases — no LLM call, no signup.

Latest benchmark · publisher runVerified

8/9/2026, 18:53 UTC · 23 cases

Lift

+83%

96% vs 13%

Agent turns

1.0 vs 1.0

Tokens

+393%

7,375.39 vs 1,497.57

fail → pass · 19pass → pass · 3fail → fail · 1

Per-case results (23)

Case	Status	Duration (ms)			Turns			Tokens			Tool calls
Case	Status	Without	With	Δ	Without	With	Δ	Without	With	Δ	Without	With	Δ
case-01 I want to generate a short video clip using two reference photos: https://example.com/boy.jpg and https://example.com/girl.jpg. Please create a video where the boy from image 1 and the girl from image 2 walk together across a sunlit campus while the camera follows them. Include background sound, set the duration to 5 seconds, and render at 720p.	fail→pass	8,218	6,923	-16%	1	1	0%	1,411	7,866	+457%	0	0	—
case-02 Please animate my character image at https://example.com/character.jpg using the movement from a reference dance clip at https://example.com/dance_routine.mp4. I want to perform a motion migration in professional mode for a 5000 millisecond duration, preserving the original audio track from the clip.	fail→pass	8,181	4,258	-48%	1	1	0%	1,505	7,246	+381%	0	0	—
case-03 I have a storyboard folder located at `./storyboards_v2` with keyframes for shots 1 through 10. Please run the batch video rendering pipeline for these shots and save all generated mp4 files and thumbnails into `./output/scene_v2`.	fail→pass	11,087	5,088	-54%	1	1	0%	2,212	7,278	+229%	0	0	—
case-04 I have two rendered MP4 files output/video_01.mp4 and output/video_02.mp4. Provide a standard terminal command using FFmpeg to concatenate these two video files seamlessly with a crossfade transition.	pass→pass	7,512	8,857	+18%	1	1	0%	1,444	8,087	+460%	0	0	—
case-05 I want to train a custom LoRA model on my local NVIDIA GPU using Kohya_ss for a dataset of 20 images in ./dataset. What terminal command or configuration file should I use to initiate the training?	pass→pass	12,657	8,212	-35%	1	1	0%	2,354	7,929	+237%	0	0	—
case-06 I need to download a video from https://www.youtube.com/watch?v=dQw4w9WgXcQ at 1080p resolution to use as a local asset. What command line tool and flags should I run?	pass→pass	7,665	4,309	-44%	1	1	0%	1,436	7,301	+408%	0	0	—
case-07 Generate a detailed 4K artwork of a cyberpunk street market at night with neon lights. I need the absolute highest quality image model available, in 16:9 aspect ratio.	fail→pass	9,021	5,563	-38%	1	1	0%	1,376	7,519	+446%	0	0	—
case-08 I need to generate a landscape photo of autumn mountains. Cost efficiency is my top priority, so pick the lowest cost image generation model in 2K resolution.	fail→pass	12,951	5,970	-54%	1	1	0%	2,274	7,526	+231%	0	0	—
case-09 I have a character design photo at https://example.com/hero.jpg and want to generate a new image of this character walking in a dark forest at night with a 16:9 ratio.	fail→pass	8,294	3,441	-59%	1	1	0%	1,438	7,051	+390%	0	0	—
case-10 Create an 8-second video of ocean waves crashing against foggy cliffs at sunrise in 16:9 format. Ensure the output contains audio sound effects.	fail→pass	8,335	6,089	-27%	1	1	0%	1,311	7,528	+474%	0	0	—
case-11 Take the single photo at https://example.com/portrait.jpg and turn it into a 5-second video of the person smiling at 1080p with default model doubao-seedance-1-5-pro-251215.	fail→pass	4,567	4,172	-9%	1	1	0%	875	7,178	+720%	0	0	—
case-12 I have three reference character images: https://example.com/c1.jpg, https://example.com/c2.jpg, and https://example.com/c3.jpg. Generate an 8-second video in 1080p 16:9 where these characters interact, using multi-image character reference.	fail→pass	10,399	6,432	-38%	1	1	0%	1,823	7,843	+330%	0	0	—
case-13 Combine image https://example.com/bg.jpg and reference video https://example.com/action.mp4 into a 720p 5-second video. Prompt: build an action sequence matching the background in image 1 and action in video 1.	fail→pass	6,128	5,186	-15%	1	1	0%	1,078	7,397	+586%	0	0	—
case-14 When running video generation I received the error 'Token has been revoked (errCode 2001)'. The image API still works. What step should I take to diagnose and resolve this token issue?	fail→pass	12,031	4,604	-62%	1	1	0%	1,922	7,232	+276%	0	0	—
case-15 Generate a 4K ultra-high-definition video of a majestic waterfall in the mountains from text description in 16:9 aspect ratio.	fail→pass	8,721	3,928	-55%	1	1	0%	1,387	6,986	+404%	0	0	—
case-16 I need to generate a 15-second cinematic video from a text description showing a starship flying through a nebula.	fail→pass	11,774	6,585	-44%	1	1	0%	1,893	7,623	+303%	0	0	—
case-17 I need to log into Neodomain API for the first time. My phone number is 13800138000. How do I request an authentication code using the CLI scripts?	fail→fail	8,743	4,655	-47%	1	1	0%	1,462	7,093	+385%	0	0	—
case-18 I submitted code 123456 for contact user@example.com using login.py --login, and it printed multiple identity choices with userId usr_998877. How do I select this identity?	fail→pass	6,816	3,535	-48%	1	1	0%	1,370	6,859	+401%	0	0	—
case-19 Generate a short video of a cat playing with yarn. I want a fast, high-cost-efficiency text-to-video generation in 720p.	fail→pass	7,165	5,092	-29%	1	1	0%	1,076	7,400	+588%	0	0	—
case-20 Transfer the dance motion from https://example.com/dance.mp4 onto the character in https://example.com/avatar.jpg using motion control in std mode.	fail→pass	4,046	6,024	+49%	1	1	0%	745	6,818	+815%	0	0	—
case-21 I have reference photos https://example.com/p1.jpg and https://example.com/p2.jpg. Produce a universal video clip quickly, prioritizing generation speed and lower cost.	fail→pass	9,924	6,062	-39%	1	1	0%	1,580	7,696	+387%	0	0	—
case-22 Generate an anime portrait image of a magic scholar girl with glowing blue eyes in 1K resolution.	fail→pass	9,616	5,340	-44%	1	1	0%	1,536	7,068	+360%	0	0	—
case-23 Generate a 16:9 8s video of a quiet library with text-to-video, reproducible with seed 424242, excluding blurry or distorted text.	fail→pass	5,693	5,390	-5%	1	1	0%	936	7,110	+660%	0	0	—

These cases were also screened by the skill’s intent / safety pass.See the Trust review ↗Reproduce this benchmark in Colab ↗

How this skill was measured

DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +83 percentage points is the difference between those two pass rates over the 23 comparable cases.

Model measured: gemini-3.6-flash
Graded by: gemini-3.6-flash
Grading method: judged
Verification: verified
Pass rate without the skill: 13%
Pass rate with the skill: 95.7%
Cases attempted: 23
Cases counted toward lift: 23
Cases that regressed: 0
Skill version measured: v1
Run completed: 8/9/2026, 18:53 UTC

Measured example — same prompt, with vs. without the skill · from the benchmark run · case case-01

Prompt

I want to generate a short video clip using two reference photos: https://example.com/boy.jpg and https://example.com/girl.jpg. Please create a video where the boy from image 1 and the girl from image 2 walk together across a sunlit campus while the camera follows them. Include background sound, set the duration to 5 seconds, and render at 720p.

Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.

Show both answers

SkillScore blends 4 signals3 of 4 complete

BenchmarkPass rate WITH skill on a verified run96

Live evalPass rate on real evaluated traces85%

AI ratingNeeds ≥10 rated traces

AdoptionBreadth of real use across orgs97

Efficiency · separate axis (at equal correctness)+0% turns+346% tokens

Shown for transparency; not credited unless a verified run shows no correctness regression.

Backed by 3 of 4 signals — each added signal makes the score more robust.

Verified on gemini-3.6-flash · re-tested 49 days ago

Other measured skills in the registry, with their headline benchmark lift.