☁️Dall_e text2Image tool

Most of ComfyUI is built around wiring an image model directly into your graph. This node does something different: it hands image generation to the LLM as a callable capability, so the agent decides on its own when a conversation needs a picture and writes its own prompt for it - no manual wiring of a generation step every time. It's DALL-E 3, wrapped as a tool.

How it works

Add it to your agent's tool list, and the LLM gains a function it can call: give it a prompt, get an image back. The three generation settings - size, quality, style - are fixed on the node ahead of time, not something the LLM chooses per-call; only the prompt itself is generated at call time. That's a deliberate simplification: you decide the output shape once, and the agent only has to get the creative part (the prompt) right.

The inputs and outputs that matter

• image_size (1024x1024 / 1792x1024 / 1024x1792, default square) - pick a portrait or landscape aspect ratio if that fits your use case better than square.
• image_quality (standard / hd, default hd) - hd costs more per DALL-E 3's own pricing; drop to standard if you're generating a lot of images and don't need the extra fidelity.
• style (vivid / natural, default natural) - vivid leans toward more hyper-real, dramatic output; natural aims for more realistic, less stylized images. This is DALL-E 3's own setting, not something the pack invented.
• base_url (default https://api.openai.com/v1/) and api_key - standard OpenAI credentials; per the README, base_url can also point at a compatible relay API as long as it still ends in /v1/.
• One output: tool - wire into your agent's tool-list input.

How to install it

• ComfyUI Manager - search "comfyui_LLM_party", install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/heshengtao/comfyui_LLM_party, then restart.

Then pip install -r requirements.txt from inside the comfyui_LLM_party folder, in your ComfyUI Python environment (python_embeded\python.exe -m pip install -r requirements.txt on portable Windows). The real prerequisite is external: an OpenAI API key with billing set up, since DALL-E 3 generations are billed per image and this node calls the real OpenAI API (or a compatible relay) directly - there's no free tier here.

Common issues & troubleshooting

Calls fail with an auth error. The node ships with a placeholder sk-XXXXX in the api_key field - if you forgot to replace it with a real key, every call fails immediately. Also worth checking: OpenAI billing needs to be active on the account the key belongs to, separate from the key just being valid.

The LLM never decides to generate an image. Like every tool in this pack, whether the agent calls this depends on its description matching the conversation - if your persona/system prompt doesn't establish that the agent can produce images, the model may just describe what an image would look like in text instead of calling the tool.

Costs adding up faster than expected. image_quality: hd and larger non-square sizes both cost more per DALL-E 3's own pricing tiers - if you're testing a lot, temporarily switching to standard and square is the cheap way to iterate before locking in final settings.

The pack won't load. Separate from this node, comfyui_LLM_party has a real, current install issue where nodes fail to import after a Manager install - usually a CUDA/PyTorch mismatch with no clear compatibility list. Check that first if dall_e_tool isn't in your node browser.