🎨 DALL-E 3

This node generates an image with OpenAI's DALL-E 3 and drops it into your ComfyUI graph. It's an API node - nothing runs on your GPU, there's no model on disk. You give it a prompt and your OpenAI key, it calls the API, an IMAGE comes back. The use case is having DALL-E 3's output available inside a workflow that's otherwise doing local SD things: generate a concept here, then upscale, mask, or composite it with your local tools downstream.

Be clear-eyed about what it is, though. DALL-E 3 is a closed API model with almost no low-level control - no ControlNet, no LoRAs, no real seed determinism, and OpenAI rewrites your prompt server-side to add detail before generating. It's great for quick ideation and hard to steer precisely. If you need control, that's what the local pipeline is for.

How it works

Your prompt (optionally translated to English first, and dynamic-prompt-expanded) goes to the OpenAI images API with your chosen size and quality. DALL-E 3 does its own prompt expansion on their end, then returns the generated image. That's why two runs of the "same" prompt can differ more than you'd expect - the model reinterprets each time.

The inputs and outputs that matter

• OpenAI_API (STRING) - your key. Paste it, or set it once in the pack's API_key.json.
• prompt (multiline STRING) - what you want.
• size - one of 1024x1024, 1024x1792, 1792x1024 (the three DALL-E 3 supports).
• quality - standard or hd. hd costs more per image.
• seed and translate (default None) round it out.

The single output is IMAGE, ready to wire onward.

How to install it

Comes with the SDVN pack. ComfyUI Manager, search SDVN_Comfy_node; or:

cd ComfyUI/custom_nodes
git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node
pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt

from the ComfyUI root, then restart. As an API node, the extra step is the key: rename API_key.json.example to API_key.json in the pack folder and fill in your OpenAI key (get one at platform.openai.com). No weights to download.

Common issues & troubleshooting

Key and billing errors. The most common failure. Confirm the key is valid, pasted without stray spaces, and that the OpenAI account has billing set up - the images API is pay-per-use and won't work on an unfunded account.

Content policy rejections. DALL-E 3 is strict, and it refuses a lot that local models happily generate. A rejected prompt returns an error, not an image - that's OpenAI's moderation, not the node breaking. This is a fundamental limit of using a hosted model rather than a local one.

Only three sizes. DALL-E 3 supports exactly the three listed shapes. There's no arbitrary resolution here; upscale locally afterward if you need bigger.

Seeds aren't reliably reproducible. The seed input exists, but DALL-E 3 doesn't give you the deterministic control a local sampler does - don't expect the same seed to reproduce an image the way it would on SD.

It ignored my language. Set the translate picker to your language so the prompt reaches the API in English; None sends it as-is.