contenido v0.4 Empezar

Ecosistema · Open source · Python 3.11+

Genera video con identidad propia,
no "slop de IA".

Un monorepo Python que fusiona cinco linajes — pipeline industrial, DAG cognitivo, editorial layer, ComfyUI multi-tenant y long-form. Desde reels de 25s hasta documentales de 60 min. Self-hosted o managed.

● MoneyPrinterTurbo ● reels-af ● corredor-content ● ComfyUI ● HKUDS/ViMax
zsh — contenido v0.4
$ contenido topic "the placebo effect" --mode premium

🎬 Generating reel — task a8c12f3
└─ DAG: hunters × 4 → critic → narrators × 3 → judge
└─ Visual: ComfyUI flux_lora_brand + Higgsfield DoP
└─ TTS: Gemini Flash, sample-accurate timing

✓ Done in 94.2s
   Output: ./output/a8c12f3/reel.mp4
   Duration: 23.4s · Cost: $0.087

# 413 tests verde · 5 linajes · 7 workflows ComfyUI

4 entry points · 18 reasoners premium · 70–110s · ~$0.08–1.20

Tests verde
413
Linajes fusionados
5
Workflows ComfyUI
7
Reasoners DAG
18
Min · long-form
5–60

Linajes — qué resuelve cada uno

Cinco proyectos open source, una sola columna vertebral.

Cada linaje aporta una capacidad específica que el resto NO tiene. No es un fork de uno con plugins; es una fusión arquitectónica deliberada con ADRs documentados.

  • 01

    Industrial-horizontal

    Origen · MoneyPrinterTurbo

    API REST + colas Redis, WebUI Streamlit, 20+ proveedores LLM, 6 motores TTS, stock footage, publicación social, multi-aspect.

    El esqueleto que permite servir muchos requests en paralelo.

  • 02

    Cognitivo-profundo

    Origen · reels-af

    DAG de 18 reasoners (hunters → critic → narrators → judge), narrativa delayed-reveal, sample-accurate TTS, word-burst karaoke.

    La diferencia entre 'genera un guión' y 'encuentra el ángulo viral'.

  • 03

    Editorial

    Origen · corredor-content

    Brand voice como código, anti-alucinación vía facts.json, gate humano plan → approve → produce, pilares rotables.

    Convierte el sistema en una redacción, no solo en un generador.

  • 04

    Visual ownership

    Origen · ComfyUI · Comfy-Org

    7 workflows pre-armados (Flux LoRA, ControlNet, IPAdapter, AnimateDiff, Inpaint, Upscale), training wizard, multi-tenant, OOM auto-retry.

    El moat real: identidad visual única vía LoRA propia.

  • 05

    Long-form video

    Origen · HKUDS/ViMax

    NovelCompressor + RAGStore + ScriptPlanner 3 actos + SceneExtractor + StoryboardArtist + VLM consistency selectors.

    Video de 5–60 min: documentales, libros animados, audiolibros.

Capacidades — 13 backends

Una matriz de trece caminos para llegar al frame correcto.

El selector híbrido decide automáticamente cuál usar según presupuesto, brand identity disponible y tipo de output. Filtra abajo para entender el espacio de decisión.

◆ ComfyUI image

Flux Basic 9:16

Duración
first frame
Costo
24 GB · ~$0
Tiempo
~18s
◆ ComfyUI image

Flux + Brand LoRA

  • Brand LoRA
Duración
first frame
Costo
24 GB · ~$0
Tiempo
~22s
◆ ComfyUI image

Flux + ControlNet (pose/depth/canny)

  • Brand LoRA
  • Layout strict
Duración
first frame
Costo
24 GB · ~$0
Tiempo
~28s
◆ ComfyUI image

SDXL + IPAdapter (style transfer)

  • Style ref
Duración
first frame
Costo
12 GB · ~$0
Tiempo
~28s

Custom: IPAdapter_plus

◆ ComfyUI video

AnimateDiff + Brand LoRA

  • Brand LoRA
Duración
16 frames
Costo
16 GB · ~$0
Tiempo
~75s

Custom: AnimateDiff-Evolved, VideoHelperSuite

◆ ComfyUI image

Inpaint + Brand LoRA

  • Brand LoRA
  • Style ref
Duración
first frame
Costo
12 GB · ~$0
Tiempo
~30s
◆ ComfyUI image

Upscale 4× + Face restore

Duración
post-process
Costo
6 GB · ~$0
Tiempo
~8s
◆ Higgsfield video

Higgsfield DoP Turbo

Duración
5s clip
Costo
~$0.20/clip
Tiempo
~30s
◆ Higgsfield image

Higgsfield Soul (character)

  • Character ✓
Duración
first frame
Costo
~$0.04/img
Tiempo
~10s
◆ Higgsfield video

Higgsfield Effects (VFX)

Duración
5s overlay
Costo
~$0.10/clip
Tiempo
~25s
◆ Stock video

Stock (Pexels / Pixabay / Coverr)

Duración
variable
Costo
$0 (free tier)
Tiempo
instant
◆ LiveAvatar video

LiveAvatar (remote_http)

  • Character ✓
Duración
audio-bound, 10000s+
Costo
~$0.05/s
Tiempo
~2× tiempo real
◆ LiveAvatar video

LiveAvatar (local_cli)

  • Character ✓
Duración
audio-bound, 10000s+
Costo
48–80 GB · ~$0.005/s (amortiz.)
Tiempo
~2× tiempo real
◆ Long-form video

Long-form Plan (book → script)

  • Character ✓
Duración
5–60 min
Costo
~$1–4
Tiempo
5–15 min
◆ Long-form video

Long-form Produce (shots → final)

  • Brand LoRA
  • Character ✓
Duración
5–60 min
Costo
~$15–80
Tiempo
30–300 min

Dos pipelines · misma columna vertebral

De 25 segundos a sesenta minutos, sin cambiar de stack.

Reels y long-form reutilizan el mismo LLM router, las mismas LoRAs ComfyUI, el mismo TTS sample-accurate y el mismo editor ffmpeg. Cambia solo la capa de orquestación narrativa. Esto NO existe en MoneyPrinterTurbo ni en ViMax por separado.

Reel · 25 segundos

DAG cognitivo · delayed-reveal

  1. 1

    Input

    topic: 'el efecto placebo'

  2. 2

    Hunters × 4

    12 candidatos paralelos (specific_figure, reversal, temporal, cross_domain)

  3. 3

    Critic + Narrators

    top 3 → 3 scripts delayed-reveal → judge elige winner

  4. 4

    Visual + Audio

    5 beats × (Higgsfield DoP + Gemini Image + Gemini Flash TTS)

  5. 5

    Stitch

    single-pass ffmpeg + word-burst libass

Wall-clock
70–110s
Costo
$0.08–1.20
Entry
topic / url / subject

● Long-form · 5–60 min

ViMax-inspired · 2 fases con gate humano

  1. 1

    Input

    novel: 50,000 palabras / script / idea

  2. 2

    Compress + RAG

    N chunks paralelo → LLM compresión → store BAAI/bge embedding

  3. 3

    Script Planner

    intent router (narrative/motion/montage) → NarrativeArc 3 actos

  4. 4

    Scenes + Storyboard

    SceneExtractor → 6-12 escenas → StoryboardArtist → 30-100 shots

  5. 5

    Shoot + Consistency

    ComfyUI + Reference selector + Best-of-N VLM × cada shot

  6. 6

    Stitch

    ffmpeg single-pass + narration + subs

Wall-clock
45–300 min
Costo
$16–80
Entry
book / script / idea

Stack compartido

core.llm_router · core.visual.generation · core.tts · core.editor · core.editorial

Apache 2.0 · GitHub público

Construye tu pipeline visual con tu propia voz.

Open source, multi-tenant, cost-tracked, con LoRA training wizard incluido. 413 tests verde. Cero vendor lock-in.

Built in Veracruz · MX · for Latam first