ComfyUI Node

create_image _O

The deprecated DALL-E node that can't talk to modern OpenAI anymore

By omar92·Created 3 years ago·Updated 5 months ago· 181
create_image _O
  • openai
  • prompt
  • IMAGE
  • MASK
number1
size256x256

This is the same DALL-E image-generation idea as the pack's create image _O, except it's the version that got left behind when OpenAI rewrote its SDK. create_image _O (underscore, deprecated) takes a prompt, hits the DALL-E 2 API, and returns an IMAGE and a MASK - but it does it through openai.Image.create, the legacy API call that stopped existing when openai-python hit 1.0.

The tell is in the inputs. Where the modern node takes a client object, this one takes an openai input of type OPENAI - produced by the pack's own deprecated loader. And its prompt is the old STR type, not the standard STRING. If you ever see this node in a loaded workflow, it's almost certainly because the graph predates the pack's OpenAI rewrite, and it's being kept alive purely so that workflow still loads.

What it was

When it worked, the flow was: load the legacy OpenAI client, pass a prompt, and openai.Image.create(prompt=..., n=..., size=...) returned image URLs that the node downloaded and converted to a torch tensor:

  • openai (OPENAI) - the deprecated client
  • prompt (STR)
  • number (INT, 1–10)
  • size - 256x256, 512x512, or 1024x1024
  • outputs: IMAGE, MASK

Same "number is ignored" habit as its successor - the source hardcodes number = 1 again. And the mask is the PNG's alpha channel inverted, or a blank 64×64 tensor when there's no alpha.

Why it's a paperweight now

OpenAI's SDK 1.0 replaced openai.Image.create with client.images.generate(). The pack auto-installs the latest openai wheel when it's missing, so any fresh install has the new SDK - and on the new SDK, that old call just throws. The pack's own pyproject is upfront about it: "Currently, this extension does not support the new OpenAI API, leading to compatibility issues." You could pin openai<1.0 in your Python environment to resurrect it, but then every other modern OpenAI integration in your ComfyUI install - there are plenty by now - breaks instead. Not a trade you want to make.

What to do

If a saved workflow drags this node in, replace it with the current create image _O. The rewire is a couple of wires: the modern version wants a client from load_openAI _O (where you set your API key and endpoint) instead of the old OPENAI object. The IMAGE/MASK outputs are the same types, so everything downstream keeps working. And before you sink time into it - the modern DALL-E node itself is a bit fragile these days, so go in with the same expectations you'd have for any paid-API node in a local-first tool.

Install

It ships in the Quality of Life Suit pack:

cd <ComfyUI root>/custom_nodes
git clone https://github.com/omar92/ComfyUI-QualityOfLifeSuit_Omar92.git

Restart ComfyUI; a browser refresh won't register new nodes. The pack auto-updates on startup via a config.json it generates, and deprecated nodes like this one are flagged for eventual removal - treat any workflow that depends on it as on borrowed time.

CategoryO/deprecated/OpenAI/Advanced/Image

Inputs (4)

NameTypeDefaultDescription
openaiOPENAI
promptSTR
numberINT11–10
sizeCOMBO256x2563 options: 256x256, 512x512, 1024x1024

Outputs (2)

NameTypeDescription
IMAGEIMAGE
MASKMASK