Updated: 2026-05

1. About This Page

As of 2026, the sheer number and variety of AI services has grown so much that beginners can easily get lost. This page outlines only the major categories. While our class will focus on the workflow from Comfy Cloud (images) to Runway (videos), understanding where each service fits into the broader landscape will help you navigate more easily.

2. Five Major Categories

Category	Examples	Input	Output
LLM (Language)	GPT-5, Claude, Gemini, Llama	Text	Text
Image-based	Stable Diffusion, Flux, Midjourney, Adobe Firefly	Text/Image	Image
Music	Suno, Adobe Firefly	Text	Music
Video	Sora, Runway, Kling, Hailuo	Text/Image	Video
Specialized	Luma AI (3D Reconstruction), Hunyuan 3D	Image/Video	3D Models

These systems work together. For example, a workflow such as “writing a story with an LLM → creating character images with image-generation tools → animating them with video-generation tools → composing background music with music-generation tools” will be commonplace by 2026.

3. Open Source vs. Closed Source

• Open-source models: Stable Diffusion series (SD 1.5, SDXL, SD3), Flux series, Wan series. Since the models themselves can be downloaded, they can be freely combined using tools like ComfyUI
• Closed-source: Midjourney, Sora 2, Nano Banana, Veo, Kling, etc. These can only be used via an API. While they offer high quality, you cannot modify their internal workings

Comfy Cloud is primarily built around open-source components, but its architecture also allows it to call closed-source components via Partner Nodes.

4. The AI Aggregator Model

A service that allows you to use multiple models from a single UI/API.

• Pollo.ai — Provides access to Sora 2, Veo, Runway, Kling, Hailuo, Pika, and many others under a single subscription. Pricing is on the higher side.
• Adobe Firefly — While it previously focused on in-house models, it is increasingly becoming an aggregator by supporting multiple models.
• Comfy Cloud Partner Nodes — A mechanism that allows external AI models to be called from within Comfy Cloud workflows.

Whether to use the “official standalone model service” or an “aggregator” depends on your use case and budget. In class, we will focus on Comfy Cloud’s built-in models (which consume fewer credits) and demonstrate Partner Nodes as needed.

5. Representative Models in Each Category (as of May 2026)

5.1 Image Models

• Stable Diffusion: SD 1.5 (lightweight, ideal for training), SDXL, SD3
• Flux: Flux dev / schnell / Flux 2 — The top choice if you’re aiming for photo-realistic quality
• Z Image Turbo — Standard on Comfy Cloud; lightweight and high-quality
• Midjourney v7 — Accessible via Discord; paid-only
• Nano Banana (Gemini-based), GPT Image 1/2 (OpenAI)

5.2 Video Models

• Sora 2 (OpenAI), Veo 3.1 (Google), Kling 2.5, Hailuo 02, Runway Gen-3 / Gen-4
• Wan 2.2 (open source, pre-installed on Comfy Cloud)
• LTX-2.3 (Comfy Cloud standard feature)

5.3 Music-related

• Suno, Adobe Firefly Audio

5.4 Specialized Systems

• Luma AI — 3D reconstruction, Gaussian splatting
• Hunyuan 3D — Built-in Comfy Cloud

6. Benchmark and Comparison Sites

• Artificial Analysis — A Cross-Comparison of Model Quality, Speed, and Pricing
• Video Generation AI Rankings: March 2025 — Summary by CGWORLD

New models are released so quickly that a comparison article from six months ago is already outdated. I check the latest trends regularly on the website mentioned above.

7. Role in the Course

• Comfy Cloud (open-source image processing): Open nodes to explore the inner workings of diffusion models → Understand how they work
• Runway (closed-source video processing): A highly polished tool focused on output → Production exercises

Our two-pronged approach aims to develop the skill to “use commercial tools while fully understanding their inner workings and recognizing their limitations.”

8. What’s Next

• External Resources — External videos and articles
• Diffusion Mechanism — An intuitive explanation of how diffusion models work
• Getting Started — How to get started with Comfy Cloud