OpenAI DALL·E 2

DALL·E 2 is the model that made image generation a household thing back in 2022, and in 2026 it is very much the grandparent of this node family. OpenAI has shipped DALL·E 3, then GPT Image 1 and 2, each one a clear step up - so the first honest piece of advice is: if you're choosing an OpenAI image node, this probably isn't the one you want. It's here because ComfyUI's partner lineup still exposes it, it's the cheapest OpenAI image option, and there is exactly one thing it can still do that's worth knowing about.

That thing is true mask-based inpainting. DALL·E 2 predates the whole "describe the edit in a sentence" era, so it works the old way: you give it an image, a mask, and a prompt, and it regenerates only the masked area. In an ecosystem where the newest models rewrite everything and call it editing, a node that only touches what you masked has a stubborn niche appeal - see the inpainting-vs-instruction debate that's been running since Kontext shipped.

The inputs that matter

• prompt - your text. For an edit, describe what should appear in the masked region.
• size - 256x256, 512x512, or 1024x1024. Tiny by modern standards; DALL·E 2's ceiling is the 1024 square.
• n - how many images to generate (1–8).
• image + mask - the editing path, and here's the gotcha: both are required together. The node errors if you supply one without the other. The mask is white = replace, black = keep, so build it from a Mask node and wire it in.
• seed - labeled "not implemented yet in backend" in its own tooltip. That's OpenAI's status, not a typo; don't rely on it for reproducibility.

Output is a single IMAGE (or n of them), wired into Preview/Save like any image.

Cost and setup

It ships with ComfyUI core, needs the usual Comfy account + credits + current ComfyUI. Pricing is small and size-scaled: about 3.4–4.2 credits per image, so roughly two cents at the top size. That's the one unambiguously nice thing about it - it's nearly free to play with.

Where it fits

Be realistic about what you're getting. DALL·E 2's quality ceiling is from 2022, its resolutions are small, and its editing is the clunky mask-and-prompt kind. If you've landed here because you want OpenAI image generation in ComfyUI, the GPT Image node (gpt-image-1/1.5/2) is the current product with current quality, and it also does mask inpainting - same trick, much better model. Keep this node in your back pocket for the two-cent experiments and the "I only want that region changed and nothing else moved" cases, and reach for its younger sibling for anything you'd actually show someone.