跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
视频与音频

ai-content-generation

When the user wants to create AI-generated image or video content: a faceless influencer or persona for TikTok and Instagram, a faceless YouTube channel, an AI clone or talking-head of themselves, or AI ad creative (UGC, product video, static images). Also use when the user mentions 'faceless influencer,' 'faceless channel,' 'AI avatar,' 'AI UGC,' 'talking head,' 'clone myself,' 'AI ad creative,' 'AI video ads,' or tools like Veo, Google Flow, Kling, Higgsfield, Seedance, Runway, ElevenLabs, HeyGen, Arcads, Nano Banana, Midjourney, or Ideogram. Gives step-by-step, copy-paste-prompt playbooks for each use case plus the realism rules that beat AI slop. For paid-ad strategy and targeting, see performance-marketing or paid-ads.

realjaymes · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add realjaymes/marketingagentskills --skill ai-content-generation

技能说明


name: ai-content-generation description: "When the user wants to create AI-generated image or video content: a faceless influencer or persona for TikTok and Instagram, a faceless YouTube channel, an AI clone or talking-head of themselves, or AI ad creative (UGC, product video, static images). Also use when the user mentions 'faceless influencer,' 'faceless channel,' 'AI avatar,' 'AI UGC,' 'talking head,' 'clone myself,' 'AI ad creative,' 'AI video ads,' or tools like Veo, Google Flow, Kling, Higgsfield, Seedance, Runway, ElevenLabs, HeyGen, Arcads, Nano Banana, Midjourney, or Ideogram. Gives step-by-step, copy-paste-prompt playbooks for each use case plus the realism rules that beat AI slop. For paid-ad strategy and targeting, see performance-marketing or paid-ads." metadata: version: 1.5.0

AI Content Generation

Help the user actually make AI image and video content, end to end, with copy-paste prompts at every step. This skill covers four use cases, each a self-contained, do-this-then-that playbook. The hard parts (keeping a character consistent and killing the AI-slop look) are baked into the prompts, so the user can follow the steps without prior experience.

The core idea

The model is the easy part. What makes AI content win is the niche, the script, keeping the character or product consistent, and beating the AI-slop look. Every playbook bakes those in. Roughly 80% of quality is set before the model runs (research, script, consistency, realism); the model is the last 20%.

Pick the use case, then open its reference

The user wants to...Use this reference
Run a recurring AI character on TikTok or Instagram, no face on camerareferences/faceless-influencer.md
Clone themselves into a talking head and generate videos from scriptsreferences/ai-clone-talking-head.md
Make paid ad creative at volume (UGC, product video, static)references/ai-ad-creative.md
Build a faceless YouTube channel (long-form and shorts)references/faceless-youtube.md

Each reference is fully self-contained: a numbered walkthrough split into one-time setup and a repeating loop, with the exact prompts inline.

How to run this skill

  1. Identify which of the four use cases the user is after (ask one question if unclear).
  2. Open the matching reference file and follow it with the user, step by step.
  3. At each tool step, give the user the copy-paste prompt from the reference, filled in with their specifics (niche, topic, product, audience).
  4. Keep one character, one voice, and one look consistent across a user's content. Consistency is the most common failure point.
  5. Apply the realism rules below to every generation prompt (they are already written into the reference prompts; do not strip them out).
  6. Before the user publishes, run the reference's pre-publish checklist.

How the tools are named (app vs model)

The names look chaotic because three different things share the spotlight: apps you chat inside, wrappers that put one interface over many models, and the models that do the generating. Once you know which is which, every new launch slots in. The model generates. The app or wrapper is just the door you walk through to reach it.

Three buckets:

  • Assistant / app: where you type. It runs models underneath. ChatGPT, Gemini, Claude.
  • Wrapper: one interface over several models, usually for cheaper volume or shot control. Google Flow, Higgsfield.
  • Model: the thing that actually generates, split by layer (text, image, video, voice). Veo 3.1, Gemini Omni, Nano Banana, Kling.
ToolTypeMade byWhat it is / what it runs
ChatGPTAssistantOpenAILLM for scripts and ideas
GeminiAssistantGoogleLLM, and the front door to Google's image (Nano Banana) and video (Omni) models
ClaudeAssistantAnthropicLLM, best for natural voice and writing
Google FlowWrapperGoogleRuns Veo 3.1 and Gemini Omni with a scene builder and shot controls. Paid (Google AI plan, credit-based). Has a QR face-scan avatar to drop yourself into any scene; work image-first to save credits
HiggsfieldWrapperHiggsfieldRuns Kling and other video models in one UI, cheap at volume
Gemini OmniVideo modelGoogleConversational generation and editing, multi-input, 1080p, ~10s clips. Now the default in the Gemini app, Flow, and YouTube Shorts
Veo 3.1Video modelGoogleCinematic, 4K, longer shots, stable API. Best for brand films and ads
KlingVideo modelKuaishouCheap volume, reached through Higgsfield
SeedanceVideo modelByteDanceLowest-cost video
Nano BananaImage modelGoogleLocks a face, edits inside an image. Lives in the Gemini app
MidjourneyImage modelMidjourneyBest aesthetic look
GPT ImageImage modelOpenAIRenders the scene and legible in-image text in one pass. Lives in ChatGPT
IdeogramImage modelIdeogramLegible text inside an image, fallback when GPT Image falls short
ElevenLabsVoice modelElevenLabsVoiceover and voice cloning
SpeechmaVoice appSpeechmaFree library voices, no cloning
HeyGenAvatar appHeyGenTalking-head avatars, clone yourself
ArcadsUGC appArcadsAI UGC actors for ads
Google VidsAll-in-one appGoogleDoc to video with stock, AI voiceover, avatars
CapCutEditorByteDanceCaptions, color, music, assembly

The big three, decoded. Each vendor has a company name, an app you chat in, and a family of models underneath. Mixing those three is most of the confusion.

  • Google: Gemini is the assistant. Nano Banana is its image model. Veo 3.1 is its cinematic video model (API, 4K, long shots). Gemini Omni is its conversational video generation-and-editing model, now default in the Gemini app, Flow, and Shorts. Veo 3.1 did not go away. It stayed on the API for high-fidelity work while Omni took over the consu
阅读完整来源 ↗