Use when generating AI images, video, or audio via fal.ai, Kling, Seedance, Suno, Midjourney. Triggers on \"ai-media-generator\", \"ai media generator\", \"gerar imagem\", \"gerar video\"
Installs into .claude/skills of the current project.
Are you the author of Ai Media Generator?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/majinmagros-ai-media-generator)
---
name: ai-media-generator
description: "Use when generating AI images, video, or audio via fal.ai, Kling, Seedance, Suno, Midjourney. Triggers on \"ai-media-generator\", \"ai media generator\", \"gerar imagem\", \"gerar video\""
---
# AI Media Generator
Generates high-quality AI media prompts and automates browser execution.
## When to Use
- Text-to-image/video/audio
- Need platform-specific prompt crafting (Kling, Seedance, Suno, etc.)
- Want automated site operation via click-protocol
## Core Workflow
1. Select platform via `references/selector.md`
2. Read platform reference (kling.md, seedance.md, suno.md, etc.)
3. Craft prompt with 5-8 high-signal tokens
4. Validate with 10-point gate (≥8) before submit
5. Automate via `automation/click-protocol.md` if requested
## References
- `references/selector.md` — platform picker
- `templates/preset-packs.md` — 30+ presets
- Full docs in original submodule commit edc8aa5
## Checklist
- [ ] Platform selected with trigger
- [ ] Prompt scored ≥8
- [ ] No generic slop tokens
## Reference Preservation (Batch 16, #47)
Prefer edit-with-reference over text-to-image from scratch: start from a
real photo, change only what was asked, keep everything else. Check
multi-turn consistency (re-edit the result 2-3x; untouched areas must not
drift) and prompt adherence on small text/layout details. Sketch-first
(rough scribble as composition guide) beats describing layout in words.
Fine details still fail (instrument strings, key counts) - zoom-verify
before delivery.
```text
Pedido: "thumbnail YouTube, neon cyberpunk, texto curto"
Plataforma: image → references/selector.md aponta Midjourney (estilo) ou fal.ai (custo)
Prompt: "neon cyberpunk portrait, rain, high contrast --ar 16:9" (6 tokens de sinal)
Gate: 9/10 → submit; se <8, refina antes de gastar crédito
```
## Prompt JSON retrato-vibe (leva YouTube rodada 10)
Constraint anti-AI-slop para personagens: sem "4K/HDR/ultra-detail"; defina cada traco (idade, pele com textura, assimetria leve, luz imperfeita); anti doll-face (sem olhos gigantes/labios inflados/simetria total). Trave em JSON versionado, nao em prompt solto.