Together Image Generator

The shortcut is the whole point. This node doesn't run a model on your machine, doesn't download gigabytes of weights, doesn't touch your VRAM. You type a prompt, hit queue, and a few seconds later an IMAGE tensor comes out the other side - generated by Black Forest Labs' FLUX.1 Schnell running on Together AI's servers, and it costs nothing.

It's a text-to-image node in the classic sense: text in, image out. Behind the scenes it calls Together's hosted image API with the black-forest-labs/FLUX.1-schnell-Free model at a fixed 4 steps - which is all Schnell needs, and the reason the free endpoint is fast and viable at all. You get the "FLUX look" without owning the hardware. The quality ceiling is a real one: this is draft-grade output, fine for storyboards, icon mockups, prompt experiments, and prototyping workflows. If you need a hero image, you're better off with a local model. If you need to know whether an idea works, this gets you there in seconds.

The inputs that matter

• prompt - the only thing you'll actually write. Multiline, and it's validated: an empty prompt is an error, not a silent no-op.
• api_key - paste a Together key here, or leave it blank and set TOGETHER_API_KEY in a .env file next to the pack. Blank input falls back to the env var.
• width / height - 512 to 2048 in steps of 256, default 1024. That's roughly FLUX's happy zone; it'll do 2048 but the free tier is more likely to shrug.
• seed - for reproducible drafts.
• num_images - 1 to 4 images per call.

Output

A single IMAGE output. Wire it into a PreviewImage or SaveImage node just like any other image - the pack's own workflow does exactly that. One honest gotcha: because the output is a proper IMAGE tensor, you can chain this straight into img2img, an upscaler, or your usual post-processing. That's the nice bit - it's not a dead-end web image, it's a first-class tensor in your graph.

Installing it

Install via ComfyUI Manager (search "Together Vision") or:

cd ComfyUI/custom_nodes
git clone https://github.com/theshubzworld/ComfyUI-TogetherVision

Then restart ComfyUI. The dependencies are light - just together and python-dotenv, both in requirements.txt, and Manager will offer to install them. There are no model downloads, which is the entire appeal of this pack. You do need a free API key from the Together AI console; the pack ships a .env template so you can drop the key in and never paste it into a node again.

Troubleshooting

• "No API key found" - the api_key input is blank and there's no .env. Fill one of them.
• "Prompt cannot be empty" and the dimension/num_images validation errors are the node's own guards; read the message and adjust.
• There's no built-in rate limiting in this node, and the free endpoint is rate-limited server-side. Fire a few, not a hundred.

One last thing worth knowing: the pack README is now just a deprecation notice pointing at the author's newer ComfyUI-NvidiaVision. This pack still works (last touched December 2025), but it's unmaintained. Fine for drafts; check the successor before you build anything permanent on it. And remember the privacy angle - every prompt leaves your machine. For private work, local generation still wins.