Image Size (Cloud Models)

The name is a lie, and that's fine. Image Size (Cloud Models) doesn't call OpenAI's API, doesn't generate anything, and doesn't need a token or a key. It's a resolution preset menu for GPT-image-1 - the cloud model Open Creative Studio hands your text to when you're running in its cloud mode instead of on your own GPU. The node's job is to be the sizing half of a workflow whose generation half lives somewhere else.

Open Creative Studio (formerly AP Workflow) is a free all-in-one ComfyUI workflow by GianoBifronte, and it pairs local models with hosted ones like GPT-image-1. This node, from the OCS_Nodes companion pack, is what sets the canvas when the actual generation happens in the cloud.

How it works

Open the dropdown and you get three choices, plus a header label so you know what you're looking at:

• 1536x1024 (3:2 | 1.6MP)
• 1024x1536 (2:3 | 1.6MP)
• 1024x1024 (1:1 | 1MP)

Those are GPT-image-1's supported output sizes, and the node turns your pick into a width, a height, a 1536x1024-style aspect string, a blank latent at that resolution, and your batch size. The code is a tiny lookup table plus a torch.zeros latent - there's no network code in the file at all. It's a presets widget that keeps your cloud calls from bouncing because you guessed a resolution the model doesn't accept.

Inputs and outputs

Only two real inputs:

• Aspect Ratio - the three presets above. There's no "custom" here, unlike the local-image sibling; cloud models define what they accept, so you don't get to improvise.
• Batch Size - 1 to 64, how many images you're asking the cloud for in one go.

Outputs: aspect_ratio (a WxH string for combo inputs), image_width, image_height, batch_size, and image_latent. Same caveat as the local node: that latent is a 4-channel zeros tensor sized for SD-style pipelines. GPT-image-1 isn't a latent diffusion model at all, so you almost certainly want the INT outputs and the aspect string, and let whatever node actually sends the cloud request worry about its own encoding.

Install

ComfyUI Manager → search "Open Creative Studio Nodes" → install, then restart. Or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/alessandroperilli/OCS_Nodes

No extra Python dependencies, no models to download. This is a UI convenience node - it costs you nothing to keep installed even if you're only running local models.

Gotchas

• It doesn't call the API. If you wire this node up expecting images, you'll get nothing. It feeds the nodes that make the actual request inside the OCS workflow.
• You can't type a custom size. If you need an unusual aspect for a cloud call, this node won't do it - that's deliberate, since GPT-image-1 rejects unsupported sizes.
• Don't bother with the latent output here. It exists so the node's output signature matches its local sibling, but a cloud call doesn't consume it.