OAI Dall_e Image🧸

This node sends your prompt to OpenAI's DALL-E 3 and drops the result back into ComfyUI as an image. It pairs naturally with Style Prompt: let the LLM write a rich narrative prompt, feed it here, get a DALL-E render, all in one graph. If you like DALL-E's look but want it living alongside your local workflows, this is the bridge.

Be realistic about what it is, though. DALL-E 3 is a closed, paid API and it's no longer the frontier. The whole reason the open-source community exists is that models like Flux caught up to and passed DALL-E on quality while running locally and free. So reach for this when you specifically want DALL-E's aesthetic or its text rendering, not as your default image engine.

How it works

You write a prompt, the node calls the OpenAI API, and a dall-e-3 image comes back. The node only supports DALL-E 3, not the older DALL-E 2, which the author dropped on purpose because the parameters differ too much and 3 is simply better.

The one behavior that surprises everyone: DALL-E rewrites your prompt. OpenAI silently expands and rephrases whatever you send, as a misuse-control measure, and generates from that. The node surfaces the rewritten version on its Dall_e_prompt output so you can see what actually got rendered. It's worth wiring up, because sometimes the rewrite explains a weird result.

The inputs that matter

• prompt (multiline): your text. Narrative, full-sentence prompts work better here than terse SD-style tags.
• image_size: 1024x1024, 1792x1024, or 1024x1792. Square or wide/tall.
• image_quality: standard or hd (default hd). HD costs more.
• style: vivid or natural (default natural). Vivid pushes saturation and drama, natural is more restrained. This is DALL-E 3's own style flag.
• batch_size (1 to 8): images per run. There's a seed field too, and although it doesn't change the render the way a KSampler seed does, setting it to randomize/increment lets the node re-run automatically on each queue.

Outputs: image, Dall_e_prompt (the rewrite), plus help and troubleshooting.

Installing it

Install Plush-for-ComfyUI via ComfyUI Manager (search "Plush", install Plush-for-ComfyUI, restart), or manually: cd ComfyUI/custom_nodes && git clone https://github.com/glibsonoran/Plush-for-ComfyUI.git, then pip install -r requirements.txt, restart. You'll need the OpenAI python library 1.3.5+, which requirements.txt handles.

Common issues

You need a paid OpenAI API key in the OPENAI_API_KEY or OAI_KEY environment variable, and setting an env var usually means rebooting the machine before it's seen. The other classic DALL-E annoyance is policy rejections: OpenAI will refuse prompts it flags, and this fires on surprisingly benign subject matter, throwing a ComfyUI error when it does. That's the API talking, not the node. The prompt rewrite is also why your output sometimes ignores a detail you clearly asked for, since the rewrite may have dropped or changed it; check the Dall_e_prompt output to see. Set a spending cap on your key, and read the troubleshooting output when a run fails.