ComfyUI-Bricolage

⚠️ This is a personal repo

This repository exists for my own ComfyUI experiments and utilities. Some nodes may be adapted from, inspired by, or lightly repackaged from other custom node packs for convenience, primarily to reduce dependency sprawl for small workflows.

• No guarantees
• No support
• No roadmap

---

Included Controls

~~DisplayString~~

~~A lightweight read-only display control for showing string output directly in the ComfyUI graph UI.~~

~~Originally adapted from rgthree's display-style controls, but simplified for my own use cases where only plain string output is needed.~~

5/21/2026:

Discovered after the fact that ComfyUI already includes a built-in Preview As Text node, which largely covers this use case.

5/22/2026:

Preview As Text node doesn't retain the text after reloading the PNG, either. Will need to take yet antoher stab at getting that to work.

---

RandomArtist

Selects a random artist name from a bundled SDXL artist list.

Supports simple filtering modes:

• any — any artist from the list
• photo — artists whose entry contains photo
• illustration — artists whose entry contains illustration

The node intentionally always reports itself as changed so a new artist is selected each execution.

---

LM Studio

A combined LLM / VLM node that talks to a locally running LM Studio server via its OpenAI-compatible API.

Inputs:

• model — the model identifier as LM Studio knows it (no default; copy it from LM Studio's model list)
• prompt — user message sent to the model
• system_prompt — optional system message prepended to the conversation
• load_model / unload_model — toggles to explicitly load and unload the model around each call, useful when VRAM is limited
• image (optional) — when wired up, the call is sent as a vision (VLM) request with the image attached; when left disconnected, a plain text (LLM) request is made instead

The server base URL defaults to http://localhost:1234. To point it at a different host or port, change LM_STUDIO_BASE_URL in src/lm_studio.py.

---

Random SDXL Latent

Generates a random latent image in one of SDXL's native aspect ratios, along with the corresponding pixel dimensions.

Outputs:

• LATENT — a random latent tensor ready to feed into a KSampler
• WIDTH — the image width in pixels (1024, 1152, 1216, 1344, or 1536)
• HEIGHT — the image height in pixels (1024, 896, 832, 768, or 640)

The node picks from all 9 supported SDXL dimensions (including portrait and landscape variants) with equal probability.

---

Text Concat Simple

Concatenates two strings: a text field and an input string.

Inputs:

• text — the static text field
• incoming_text — the incoming string from another node
• incoming_first — when on, puts the incoming string before the text field; when off, puts it after

Output: the concatenated result.

---

Simple Wildcards

Expands wildcards in the ComfyUI standard format: { item_a | item_b | item_c }.

Input: multiline text with zero or more wildcard expressions.

For each wildcard found, one item is chosen at random and replaces the entire { ... } block. Multiple wildcards in the same text are each resolved independently (but are not nested).

Example:

• Input: a { red | blue } ball in a { big | small } room
• Output: a blue ball in a big room (or any other combination)