EcoHash Image Generate

The honest pitch up front: you don't run anything here. EcoHashImageGenerate is one of six nodes in the ComfyUI-EcoHash pack, and every one of them is a thin client for the EcoHash hosted API. You type a prompt, the node POSTs it to a server, and a finished IMAGE tensor lands back in your graph as if a local sampler had made it. No 20 GB checkpoint download, no VRAM, no torch version war, no VAE hunt. That's the whole draw - and the whole trade, because it costs per call, your prompt leaves your machine, and there's a content filter on the far end you can't argue with - the standard API-node bargain, documented well in the community's own postmortems of hosted-wrapper packs.

Why you'd reach for it

Three good reasons. You have no GPU (or one too small to care about) and want the whole workflow cloud-side. You have a GPU but want to compare hosted models next to your local ones in the same graph. Or you're drafting ideas fast and don't want to babysit a sampler. Honestly: if you can run Z-Image Turbo locally, it's free after electricity and this node isn't - the API wins when you can't run it. There's also a free starter credit on signup, so the "is this worth it" test costs you nothing.

How it works

The model dropdown is populated live from EcoHash's model catalog (api.ecohash.com/platform/models), cached for five minutes, with a packaged snapshot fallback if the API is unreachable - so the graph still loads when the catalog is down. On queue, the node POSTs to /images/generations with your model, prompt, and size, asks for a base64 image back, and decodes it into a standard ComfyUI IMAGE. There are no weights on disk anywhere in this pack; requirements.txt is literally one line, requests.

The inputs that matter

• model - right now three options: z-image-turbo (Alibaba's fast 8-step distilled model and the sensible default - a big chunk of the ecosystem runs on this one), qwen-image (better with text in the image), and flux2-klein (BFL's small Flux). The list changes as the catalog changes, so don't be surprised when it grows.
• prompt - the whole job. Multiline.
• size - 1024x1024 default, plus 1024x768, 768x1024, 768x768, 512x512. No arbitrary sizes here.
• steps - 0 means "model default." Leave it at 0.
• seed - -1 is random. Set a real value when you want to reproduce an output.

The single output is IMAGE, which wires straight into PreviewImage or SaveImage - or into another EcoHash node, like Image Edit.

Where people get burned

• ComfyUI caches node outputs. Re-queueing an identical graph returns the cached result with no API call and no new image. Change any input - bump the seed - to force a fresh generation. This trips everyone once.
• 401 = invalid API key, 402 = your account ran out of credit. The pack deliberately has no key widget; the key comes from the ECOHASH_API_KEY env var or a config.ini (see below), because a key typed into a node would get saved into every workflow JSON you share.
• (catalog unavailable) in the model dropdown means the live fetch failed and the snapshot fell through too. Restart ComfyUI.
• One quirk of the underlying model, not the node: z-image-turbo is a distilled model and shows low seed-to-seed variance - different seeds, same prompt, near-identical images. If you want variety, change the prompt too.

Install

ComfyUI Manager: search EcoHash, install "ComfyUI-EcoHash", restart. Or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/ecohash-ai/ComfyUI-EcoHash.git
cd ComfyUI-EcoHash
pip install -r requirements.txt

Then set your key (sign up at ecohash.com for the free starter credit) either in the environment or in config.ini:

export ECOHASH_API_KEY="eco_..."
# or copy config.ini.example to config.ini and paste the key there

That's it. No model files, no HF download, no dependencies beyond requests. The pack is brand new with no community track record yet, so the usual fresh-node advice applies - it's open source and MIT-licensed, but read what it does before trusting it with a real key.