Stillfront-comfyui-sf-nodes/README.md

SF ComfyUI Nodes

A collection of custom ComfyUI nodes developed by Stillfront. All nodes are prefixed with SF for easy searching in ComfyUI.

---

Installation

Via ComfyUI Manager (recommended)

Search for "SF ComfyUI Nodes" in the ComfyUI Nodes Manager and click Install.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/Stillfront/comfyui-sf-nodes.git
cd comfyui-sf-nodes
pip install -r requirements.txt

---

Configuration

Copy config.ini.tmp to config.ini and fill in your API keys:

[API]
wavespeed_api_key = YOUR_WAVESPEED_API_KEY

Alternatively, set keys via environment variables:

Variable	Used by
WAVESPEED_API_KEY	All WaveSpeed nodes
ANTHROPIC_API_KEY	SF LLM Chat (Claude)
GEMINI_API_KEY	SF LLM Chat (Gemini)
OPENAI_API_KEY	SF LLM Chat (GPT)
Google Application Default Credentials	All VertexAI nodes

For VertexAI nodes, authenticate via the Google Cloud CLI:

gcloud auth application-default login

---

Node Categories

Stillfront/LLM

Chat with large language models directly inside your ComfyUI workflows, with optional image input.

Node	Description
SF LLM Chat	Multi-model chat supporting Claude (Opus, Sonnet, Haiku), Gemini 3 (Pro, Flash), and OpenAI GPT-5 series. Supports image input and reports token usage.
SF Text Analyzer	Analyze and process text strings within workflows.
SF Claude Code	Send text and images to Claude through the Claude Code CLI on your own machine, using your existing Claude login instead of an API key. Image slots appear as you fill them.
SF Claude Asset Director	Point Claude at an asset library and it chooses which references suit the brief and writes the image prompt to go with them. Outputs an IMAGE batch plus the prompt, ready to wire into any generation node.

| SF Claude Library Builder | Point Claude at a folder of game assets and it writes a searchable manifest describing each one. Can also reorganise and rename the files. Feeds SF Claude Asset Director. |

The asset library workflow

These two nodes are a pair. Library Builder catalogues your assets once; Asset Director then searches that catalogue on every run without ever opening an image.

   one time, when assets change          every generation
┌──────────────────────────────┐   ┌──────────────────────────────────┐
│ SF Claude Library Builder    │   │ SF Claude Asset Director         │
│ opens every image,           │──►│ searches the manifest,           │──► images + prompt
│ writes manifest.json         │   │ opens nothing                    │
└──────────────────────────────┘   └──────────────────────────────────┘

That split is the point: cataloguing is slow (~5s per asset) but happens rarely, while choosing is instant and happens constantly.

The manifest schema (sf-asset-library/1):

{
  "schema": "sf-asset-library/1",
  "project": "Fantasy RPG",
  "art_style": "Hand-painted 2D, saturated palette, thick outlines, 3/4 view",
  "generated": "2026-10-08",
  "assets": [
    {
      "file": "characters/heroes/knight_aldric.png",
      "name": "Aldric the Knight",
      "category": "character",
      "subcategory": "hero",
      "archetype": "tank",
      "tags": ["armoured", "sword", "front view", "heroic"],
      "desc": "Armoured human knight in silver plate with a red tabard, standing front-on."
    }
  ]
}

It's shaped for fast searching rather than for humans: one flat list (cheaper to scan than a tree), the same keys on every entry, faceted fields (category / subcategory / archetype / tags) to narrow by, and a one-sentence desc carrying the meaning. The top-level art_style matters more than it looks — Asset Director reads it and keeps every prompt it writes consistent with it, so a library generates in one coherent style.

Library Builder modes:

Mode	What it touches
manifest only (default)	Nothing. Reads the images, returns the JSON as text
organise into a copy	Writes a tidy, renamed copy to output_path; originals untouched
organise in place	Renames and moves your actual files

write_manifest is off by default, so the first run shows you the JSON without writing anything. Turn it on once you're happy, or paste the text in yourself.

Start in manifest only. Reorganising is genuinely useful on a folder of IMG_0021.png and final_v2 copy.png, but it moves real files — try it on a copy first.

Large libraries are processed in batches, not one session. batch_size (default 15) caps how many assets Claude handles per call — asking it to hold a whole large library open, compose one giant manifest, and write the file all in a single uninterrupted session is what used to blow up in turns and cost on libraries of 100+ assets, with nothing produced at the end. Each batch now only reads its own assets and replies with JSON; the node assembles and writes the manifest itself, so no file-writing tool is ever handed to Claude for this.

A hidden checkpoint (.sf_library_builder_checkpoint.json, removed automatically on success) tracks which assets are already catalogued. If a run fails or times out partway, rerunning with the same settings resumes — finished batches are skipped, not re-paid for.

If a batch's reply comes back cut off — the model ran past its reply length limit partway through describing a wordy set of assets — that batch is automatically retried as two smaller halves, repeatedly if needed, rather than failing the run. You'll see it in the log as Batch 7/7: the reply was cut off… Retrying as two batches of 5 and 6.

Using SF Claude Asset Director

Give it a folder of assets and a brief; it returns the references Claude chose and the prompt it wrote for them. Wire images and prompt straight into Nano Banana, Seedream, or anything else that takes an image batch and a text prompt.

It decides from a manifest, not from the pixels. Put a file in the library — assets.json by default, though any text or JSON works — describing what each asset is:

{"assets": [
  {"file": "knight.png", "desc": "Armoured knight character, heroic, front view"},
  {"file": "castle.png", "desc": "Stone castle environment, background location"}
]}

That keeps a large library cheap to work with: no image is opened unless you ask for it.

Input	Purpose
library_path	The asset folder. Point it at a Google Drive for Desktop sync folder to work off Drive
manifest	Descriptions file inside the library (default assets.json); falls back to filenames if absent
system_prompt	Standing rules — e.g. "Choose only character assets unless a background is asked for"
instruction	The brief for this run. Type it or wire a string in
max_images	Ceiling on how many references Claude may pick
pad_to_fit	On (default): scale each image onto a shared canvas, keeping aspect ratios. Off: stretch them all to one size
look_at_images	Off (default). On, Claude opens the files as well — better picks on an unlabelled library, but ~5s per image
seed	Not sent to Claude; change it to force a fresh choice rather than ComfyUI's cached one

Outputs are images (batch), prompt, and picked (the filenames chosen, for checking its work).

Why pad_to_fit exists: a ComfyUI image batch is one tensor, so every frame must share dimensions — and an asset library won't. Padding preserves aspect ratios at the cost of black bars; stretching avoids bars at the cost of distortion. Which is better depends on the model you're feeding.

Requires Claude Code installed and signed in, same as SF Claude Code.

Using SF Claude Code

This node shells out to the Claude Code CLI, so it bills against your Claude subscription rather than an Anthropic API key.

Requirements: Claude Code must be installed and signed in on the machine running ComfyUI. The node searches your PATH plus the usual install locations, and gives a clear error if it can't find it.

Input	Purpose
prompt	What you want Claude to do
system_prompt	Shapes the reply, e.g. "Reply with only the prompt text, no preamble." Leave blank to skip
model	default, opus, sonnet, or haiku
timeout_seconds	Give up after this long (default 300)
seed	Not sent to Claude — change it to force a fresh reply instead of ComfyUI's cached one
image_1…	Connect an image and the next slot appears, up to 16

Worth knowing before you build a big batch around it:

• Roughly 5–6 seconds per call minimum. Claude Code is an agent that reads each image as a tool call, so it's slower than a direct vision API request.
• Subscription rate limits apply. Heavy batch use will hit them; SF LLM Chat with an API key is the pay-per-token alternative that won't.
• Not portable. Anyone else installing this pack needs their own Claude Code install and login for this particular node. Every other node in the pack is unaffected.

---

Stillfront/VertexAI

Generate images and videos using Google Cloud Vertex AI. Requires a Google Cloud project and Application Default Credentials.

Node	Description
SF VertexAI Imagen 3 Text to Image	Generate images from text using Google Imagen 3.0. Supports multiple aspect ratios, safety settings, and up to 4 images per run.
SF VertexAI Imagen 4 Text to Image	Generate images using the latest Imagen 4.0 model.
SF VertexAI Imagen 4 Upscale	Upscale images using Imagen 4.0.
SF VertexAI Nano Banana Pro	Image generation via the Nano Banana Pro model on Vertex AI.
SF VertexAI Nano Banana Pro Edit	Edit existing images using Nano Banana Pro on Vertex AI.
SF VertexAI Veo 3.1 Text to Video	Generate videos from text prompts using Google VEO 3.1 via Vertex AI.
SF VertexAI Veo 3.1 Image to Video	Animate a still image into a video using Google VEO 3.1 via Vertex AI.

---

Stillfront/WaveSpeed

Image and video generation nodes powered by the WaveSpeed AI API. All nodes require the SF WaveSpeed Client node to authenticate.

Setup

Connect an SF WaveSpeed Client node (enter your WaveSpeed API key) to any WaveSpeed generation node.

Video Generation

Node	Description
SF WaveSpeed VEO 3.1 Text to Video	Google VEO 3.1 text-to-video generation.
SF WaveSpeed VEO 3.1 Image to Video	Google VEO 3.1 image-to-video animation.
SF WaveSpeed VEO 3.1 Reference to Video	VEO 3.1 video generation using a reference image.
SF WaveSpeed VEO 3.1 Fast Text to Video	Faster VEO 3.1 text-to-video variant.
SF WaveSpeed VEO 3.1 Fast Image to Video	Faster VEO 3.1 image-to-video variant.
SF WaveSpeed Sora 2 Text to Video	OpenAI Sora 2 text-to-video generation.
SF WaveSpeed Sora 2 Text to Video Pro	Sora 2 Pro — higher quality text-to-video.
SF WaveSpeed Sora 2 Image to Video	OpenAI Sora 2 image-to-video animation.
SF WaveSpeed Sora 2 Image to Video Pro	Sora 2 Pro — higher quality image-to-video.
SF WaveSpeed Wan 2.5 Text to Video	WAN 2.5 text-to-video generation.
SF WaveSpeed Wan 2.5 Text to Video Fast	WAN 2.5 faster text-to-video variant.
SF WaveSpeed Wan 2.5 Image to Video	WAN 2.5 image-to-video animation.
SF WaveSpeed Wan 2.5 Image to Video Fast	WAN 2.5 faster image-to-video variant.
SF WaveSpeed Wan 2.2 Animate	WAN 2.2 image animation.
SF WaveSpeed InfiniteTalk	Animate a portrait image with audio to create a talking-head video.
SF WaveSpeed InfiniteTalk Multi	InfiniteTalk with multiple reference inputs.

Image Generation

Node	Description
SF WaveSpeed Nano Banana Text to Image	Nano Banana text-to-image generation.
SF WaveSpeed Nano Banana Edit	Edit images using the Nano Banana model.
SF WaveSpeed Nano Banana Pro Text to Image	Nano Banana Pro text-to-image.
SF WaveSpeed Nano Banana Pro Text to Image Multi	Nano Banana Pro with multiple inputs.
SF WaveSpeed Nano Banana Pro Text to Image Ultra	Nano Banana Pro Ultra quality variant.
SF WaveSpeed Nano Banana Pro Edit	Edit images with Nano Banana Pro.
SF WaveSpeed Nano Banana Pro Edit Multi	Nano Banana Pro edit with multiple inputs.
SF WaveSpeed Nano Banana Pro Edit Ultra	Nano Banana Pro edit Ultra quality variant.
SF WaveSpeed Qwen Text to Image	Qwen text-to-image generation.
SF WaveSpeed Qwen Edit	Edit images using Qwen.
SF WaveSpeed Qwen Edit Plus	Qwen Edit with extended capabilities.
SF WaveSpeed Qwen Edit LoRA	Qwen edit with LoRA support.
SF WaveSpeed Qwen Edit Plus LoRA	Qwen Edit Plus with LoRA support.
SF WaveSpeed Seedream V4	ByteDance Seedream V4 image generation.
SF WaveSpeed Seedream V4 Sequential	Seedream V4 sequential generation.
SF WaveSpeed Seedream V4 Edit	Edit images with Seedream V4.
SF WaveSpeed Seedream V4 Edit Sequential	Seedream V4 sequential editing.
SF WaveSpeed Wan 2.5 Text to Image	WAN 2.5 text-to-image generation.
SF WaveSpeed Wan 2.5 Image Edit	Edit images using WAN 2.5.
SF WaveSpeed Flux Kontext Dev	Flux Kontext Dev image generation.
SF WaveSpeed Flux Kontext Pro	Flux Kontext Pro image generation.
SF WaveSpeed Flux Kontext Max	Flux Kontext Max — highest quality variant.
SF WaveSpeed Flux ControlNet Union Pro 2	Flux image generation with ControlNet guidance.

Upscalers

Node	Description
SF WaveSpeed Image Upscaler	Upscale images via WaveSpeed AI.
SF WaveSpeed Runway Upscale	Upscale images using RunwayML via WaveSpeed.

---

Stillfront/Utils

Utility nodes for building and enhancing workflows.

Node	Description
SF Prompt List	Combine 2–50 text inputs into a single comma-separated string. Add or remove inputs using the inputcount control.
SF Qwen Resolution	Visual resolution preset selector with aspect ratio preview.
SF Text Analyzer	Analyze text content within a workflow.
SF Google Sheet Cell	Read a single cell from a publicly shared Google Sheet. Specify the sheet tab by GID, row by number (1-indexed), and column by letter (A, B, C, ...).
SF Google Sheet Range	Read a range of cells from a publicly shared Google Sheet and join them into a single string with a configurable delimiter. Supports multi-row, multi-column ranges.
SF Google Sheet Prompt Queue	Run a batch of prompts straight from a Google Sheet. Reads prompts from one column, skips any row you've marked as finished in a second column, and sends the rest downstream — the graph runs once per prompt.

Google Sheet nodes require the sheet to be shared as "Anyone with the link can view." The sheet tab GID is found in the browser URL after #gid= when you have that tab open.

Using SF Google Sheet Prompt Queue

Set your sheet up with prompts in one column and an empty column beside it for tracking:

	A	B
1	Prompt	Status
2	a red cat on a windowsill	done
3	a blue dog in the snow
4	a green bird mid-flight

With the defaults (prompt_col A, status_col B, skip_first_row on), a run sends rows 3 and 4 and skips row 2. Each prompt triggers a separate pass through the downstream graph.

• max_rows caps how many prompts go out per run — 10 by default.
• Marking is manual. Type anything into the status column — done, x, executed — and that row is skipped next time. Clear the cell and it gets queued again.
• Scanning stops at the first empty prompt cell, so anything below a gap is ignored.
• The sheet is re-read on every run, so edits you make in the browser are picked up automatically. Add rows or clear status cells and just hit Run again.
• Refresh sheet is optional — it reports how many prompts are pending without running the graph.

---

Requirements

• Python 3.10+
• ComfyUI
• See requirements.txt for full Python dependency list

License

MIT License — see LICENSE for details.