Trends

.ai

Guías

los mejores modelos de IA de código abierto

Cómo instalar ComfyUI

The node-based powerhouse for Stable Diffusion, Flux, Wan video and almost every new open source model. Free, open source, for Windows, Mac and Linux.

Generación de imágenes

Windows

Mac

Linux

Licencia: GPL-3.0

Sitio oficial

GitHub

Documentación

Guía de nodos de ComfyUI y solución de problemas

Cada nodo popular explicado, con sus errores y cómo arreglarlos.

¿Qué es ComfyUI?

ComfyUI is a free, open source program for running AI image and video models on your own computer. Instead of one fixed form, it shows your generation as a graph of connected nodes: a model loader, a text encoder, a sampler, a decoder and so on. Each node does one job, and you wire them together into a workflow. That sounds technical, and it is, but it is also why ComfyUI is usually the first app to support a new model: the pieces can be rearranged for anything.

It is the right choice if you want the newest models (Flux, Flux 2, Wan, Hunyuan, Qwen Image, Chroma, LTX), want to run LoRAs, ControlNets, upscalers and video in one place, or want to share exact, reproducible workflows with others. The ComfyUI Desktop app makes the first install easy; the classic portable build gives full control. If you only want to type a prompt and get a picture, Fooocus or Invoke are gentler starts.

Sin instalación: ejecuta modelos de IA en la nube

BitVector Prism es la forma más fácil de empezar: elige un modelo, escribe un prompt y genera en una aplicación web limpia, sin configurar nada. BitVector también está disponible en Discord y en la web (SpyGlass).

Probar BitVector

Qué necesitas

Windows 10 or 11 with an NVIDIA graphics card (6 GB VRAM for SDXL, 12 GB or more for Flux and video; 24 GB is comfortable) is the best supported setup. AMD cards work on Linux with ROCm and on Windows through DirectML or ZLUDA, with more effort. Apple Silicon Macs (M1 and later) run ComfyUI natively through PyTorch MPS; 16 GB of unified memory is the practical minimum, 32 GB or more for Flux. You also need 20 to 100 GB of free disk space for model files, since one Flux checkpoint alone is about 12 to 24 GB.

Cómo conseguir e instalar ComfyUI

Open the official site (comfy.org) and download ComfyUI Desktop for Windows or Mac. It is a normal installer that sets up Python and everything else for you.
Windows alternative: download the portable ZIP from the GitHub releases page, extract it to a short path such as C:\ComfyUI, and start run_nvidia_gpu.bat (or run_cpu.bat without a GPU).
Linux, or manual install on any system: clone the GitHub repository, create a Python virtual environment, install PyTorch for your GPU (CUDA, ROCm or MPS) and then run pip install -r requirements.txt.
Download at least one model file: for example an SDXL checkpoint, or Flux (the unet file plus the clip_l, t5xxl text encoders and the ae VAE). Civitai and Hugging Face are the usual sources.
Put the files in the models folder: checkpoints go in models/checkpoints, Flux and video diffusion models in models/diffusion_models (or models/unet), text encoders in models/text_encoders (models/clip), VAEs in models/vae and LoRAs in models/loras.
Start ComfyUI. The browser opens at http://127.0.0.1:8188 (the Desktop app has its own window). Install ComfyUI Manager from the menu if it is not already there; it finds missing custom nodes and models for you.

Descarga solo desde los enlaces oficiales de esta página. Los instaladores de otros sitios pueden estar desactualizados o manipulados. ModelTrends.ai no aloja archivos.

Sin instalación: ejecuta modelos de IA en la nube

PirateDiffusion es solo para Telegram y está pensado para profesionales: miles de modelos, LoRAs y workflows mediante comandos de chat, con generación ilimitada a precio fijo.

Probar PirateDiffusion

Cómo usar ComfyUI

Load a template: the menu (Workflow > Browse Templates) has ready-made graphs for text to image with SD 1.5, SDXL, Flux and the video models. Pick one that matches a model you downloaded.
In the Load Checkpoint (or Load Diffusion Model) node, choose your model from the list. If the list is empty, the file is in the wrong folder; press R to refresh after moving it.
Type your prompt in the positive CLIP Text Encode node and anything you do not want in the negative one (Flux ignores negatives). Set the width and height in the Empty Latent Image node.
In the KSampler node set the steps (20 to 30 for SDXL, 20 for Flux dev, 4 for Flux schnell and other distilled models), the CFG (5 to 7 for SDXL, 1 for Flux) and the sampler (euler or dpmpp_2m are safe choices).
Press Run (or Ctrl+Enter). The image appears in the Save Image node and is also written to the output folder. Change the seed to get a different picture with the same prompt.
To add a LoRA, add a Load LoRA node between the checkpoint and the text encoders, choose the file and keep the strength near 0.7 to 1.0. Every model page on ModelTrends.ai lists the trigger words to put in your prompt.

Consejos y problemas frecuentes

Out of memory? Start with --lowvram, pick an fp8 or GGUF version of the model, lower the resolution or close other GPU programs. Flux fp8 fits in 12 GB; GGUF Q4 versions fit in 8 GB.
Red nodes after loading a workflow mean a missing custom node. Open ComfyUI Manager, choose Install Missing Custom Nodes, then restart.
Drag any image generated by ComfyUI back onto the canvas to restore the exact workflow that made it, including prompt, seed and model.
Keep the models folder outside the ComfyUI folder and point to it with extra_model_paths.yaml, so you can update or reinstall ComfyUI without moving dozens of gigabytes.
On a Mac, the first generation after starting is slow while shaders compile; the second one is much faster. Keep the image size at or below 1024 on 16 GB machines.

Modelos que puedes ejecutar en ComfyUI

ComfyUI runs nearly every open source model in this catalog: Stable Diffusion 1.5 and SDXL checkpoints, Flux and Flux 2, Qwen Image, Chroma, Z-Image, the Hunyuan, Wan and LTX video models, and their LoRAs. Check a model page here for the base model and trigger words, then put the file in the matching folder.

Modelos Flux

Modelos Flux 2 / Klein

Modelos Stable Diffusion XL 1.0

Modelos Stable Diffusion 1.5

Sin instalación: ejecuta modelos de IA en la nube

Sin GPU, sin descargas, sin Python. Estos servicios ejecutan los mismos modelos de código abierto en sus propios servidores, así que puedes empezar a generar desde el navegador, Discord o Telegram en un minuto.

Probar BitVector

Probar PirateDiffusion

Enlaces

Estos enlaces llevan a los sitios de sus creadores. ModelTrends.ai no está afiliado a ellos y no aloja software.

Otras apps locales

Cómo instalar Stable Diffusion WebUI (AUTOMATIC1111)

The classic Stable Diffusion web interface: tabs, sliders and thousands of extensions. The app most tutorials were written for.