ComfyUI_LiteLLM

Force the LLM into a JSON schema, no matter how much it wants to ramble

The completion node that keeps going until the answer is actually good

The tiny node that merges two conversations into one

Make the LLM read its own answer and try again, until the answer is better

Wrap any callable into a reflection filter for the completion loop

Feed a whole document to an LLM that can only swallow one chunk at a time

A filter for code-generating loops

Preview the HTML an LLM just wrote, right inside the node

Rendered HTML that comes back as an actual image in your pipeline

Serve the HTML an LLM made, and hand you a link to open it

Turn a plain list into the message format the LLM nodes understand

One node to call basically any LLM from your ComfyUI graph

Turn a list of prompts into a list of completions in one run

The completion node with a system-prompt slot bolted on front

The node that hands the rest of the pack a callable 'ask the LLM' function

The completion node that can also caption your images

Ask the LLM, look at its answer, and ask again until it's right

Point the whole pack at any OpenAI-compatible API, including ones you run yourself

A captioning function you can hand any image from your graph

LiteLLMMessage builds the conversation

LiteLLMModelProvider is the model dropdown, warts and all

The model picker that tells you what you'll pay before you spend a cent

LLLMInterpretUrl grabs text and links for you

MarkdownNode renders Markdown while passing text through

MessagesToList strips the type, keeps the data

MessagesToText turns chat history into plain text

ModifyModelKwargs tweaks your model call

PDFToImageNode turns pages into images

ShowLastMessage shows the tail of your chat

ShowMessages prints every message on the node

TextToMessages parses text into chat messages