Imagen4 Preview (fal)

Imagen4Preview is about as simple as a ComfyUI node gets: one prompt box, one image out. It calls Google's Imagen 4 (the preview tier) through fal - closed model, hosted on Google's side, nothing local - and gives you back a native ComfyUI IMAGE. If you just want a high-quality, photoreal, text-capable image and don't want to fiddle with samplers, schedulers, and a dozen sliders, that's the pitch.

It's part of gokayfem's ComfyUI-fal-API pack, which wraps fal's whole catalog behind one API key. Imagen has real pedigree, too - it's Google's flagship text-to-image line, and the researchers who built the original Imagen went on to found Ideogram.

How it works

There's no diffusion happening on your machine. The node packages your prompt into a fal request, fal calls Google's Imagen 4 endpoint, and the finished image comes back. You pay per image and wait on network latency instead of GPU time. "Preview" just means you're on the early-access version of the endpoint.

The inputs and outputs that matter

This one is refreshingly bare:

• prompt - the only input. Write a clear, natural-language description. Imagen handles full sentences and does well with legible text and photorealism, so describe the scene the way you'd describe it to a person.

Everything else - resolution, aspect ratio, sampling - is decided server-side with Google's defaults. The single output is an IMAGE, ready to wire into a Preview or Save node. If you later want more control (aspect ratios, multiple images), other nodes in the pack expose richer parameters; this one is deliberately minimal.

How to install it

Via ComfyUI Manager: search ComfyUI-fal-API, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI-fal-API.git
pip install -r ComfyUI-fal-API/requirements.txt

No model files download. Get a fal API key from fal.ai, put it in config.ini under [API] (copy config.ini.example first) or export FAL_KEY=your_key, then restart. The node shows up under FAL.

Common issues & troubleshooting

• Nothing runs / auth error. FAL_KEY isn't set. Add it and restart.
• A prompt gets rejected. Google's content filtering runs on their side, and the pack raises fal's real error message - a policy rejection will say so. Rephrase.
• Fewer knobs than you wanted. By design - this node is just a prompt. Reach for another image node in the pack if you need aspect ratio or batch controls.
• Cost. Every generation is a paid API call. The pack caches identical calls (free on repeat; force_rerun bypasses) and offers a spend guard to cap a session.
• Windows Portable. If fal-client is missing: python_embeded\python.exe -m pip install fal-client.

Use it as your "just give me a good image, quickly" button. When you need fine control or want to iterate cheaply, that's what your local models and the more configurable nodes are for.