Extensions/comfyui-bulk-prompt
ComfyUI Extension

comfyui-bulk-prompt

Batch-run your workflow once per prompt from a CSV file, Google Sheets URL, or pasted text, and optionally improve each prompt with a local Ollama model.

By sifatrads·Created 3 months ago·Updated 3 months ago· 0
sifatrads/comfyui-bulk-prompt
Nodes—
On cloudLocal install
Stars0
Updated3 months ago
Readme

ComfyUI-BulkPrompt

Batch-run your workflow once per prompt from a CSV file, a Google Sheets URL, or pasted text, and optionally improve each prompt with a local Ollama model.

The core loader has no Python dependencies — the CSV / Sheets / Paste nodes use only the standard library (csv, urllib, json). The optional 🦙 Ollama Improver node uses the official ollama Python library (pip install ollama) to talk to a running Ollama server.

Installation

  1. Clone (or copy) this repo into your ComfyUI custom_nodes folder: 
    cd ComfyUI/custom_nodes
git clone https://github.com/sifatrads/comfyui-bulk-prompt.git
  2. Restart ComfyUI.
  3. The nodes appear under the category BulkPrompt in the node menu.

The core nodes need no pip install. The optional 🦙 Ollama Improver node needs the ollama package (pip install ollama, or it is installed automatically from requirements.txt by ComfyUI Manager) plus a running Ollama server (see below).

Nodes included

| Node | What it does | |------|-------------| | 📋 Bulk Prompt Loader (CSV / Sheets) | Main node — loads one row per queue run from a CSV file, Google Sheets URL, or pasted text | | 🦙 Bulk Prompt Ollama Improver | Rewrites each row's prompt with a local Ollama model (optional, via the official ollama Python library) | | 🌐 Google Sheets Fetcher | Fetches raw CSV text from a published Sheets URL | | 🔄 Bulk Prompt Reset Counter | Resets the row counter back to 0 |

How to use

Option A — CSV File

  1. Put your .csv file inside:

    ComfyUI-BulkPrompt/csv_files/your_prompts.csv

  2. CSV format with headers (recommended):

    positive,negative,filename_tag
a red apple on a wooden table,blurry bad quality,apple
a blue car in the rain,ugly watermark,car

  3. CSV format without headers (also works):

    a red apple on a wooden table
a blue car in the rain

  4. In ComfyUI:

    • Add Bulk Prompt Loader node
    • Set source → CSV File
    • Select your file from the dropdown
    • Connect positive → CLIP Text Encode
    • Connect negative → CLIP Text Encode (negative)
    • Leave auto_loop → enabled
    • Click Queue Prompt once → the node auto-queues the next run for each row and stops at the last one (the on-node progress bar tracks it). Set loop_forever → yes to start over after the last row.

The auto-loop is driven from the browser, so keep the ComfyUI tab open while it runs. The row counter is saved in state.json, so it resumes if interrupted.

Option B — Google Sheets

  1. Open your Google Sheet
  2. File → Share → Publish to web
  3. Choose Comma-separated values (.csv) → click Publish
  4. Copy the URL (looks like: https://docs.google.com/spreadsheets/.../pub?output=csv)
  5. In ComfyUI:
    • Set source → Google Sheets URL
    • Paste the URL into sheets_url
    • Works the same as CSV from here

Option C — Paste Text

  1. Set source → Paste Text
  2. Paste your prompts into the pasted_data box. The format is auto-detected:
    • One prompt per line: 
      a red apple on a wooden table
a blue car in the rain
    • A comma-separated list on one line (each item becomes one prompt): 
      a red apple, a blue car, a green tree
    • Full CSV with a header row (mapped to the column outputs): 
      positive,negative,filename_tag
a red apple,blurry,apple
a blue car,watermark,car
  3. Queue once — it loops through every pasted row just like the CSV source. Editing the pasted text starts a fresh run from row 0.

Improve prompts with Ollama (optional)

The 🦙 Bulk Prompt Ollama Improver node rewrites each prompt with a local Ollama model before it reaches your sampler.

It uses the official ollama Python library. Install it with pip install ollama (or let ComfyUI Manager pull it from requirements.txt) and make sure your Ollama server is running (ollama serve).

  1. Add the 🦙 Bulk Prompt Ollama Improver node.
  2. Wire the loader's positive output into the node's positive input.
  3. Set url and click 🔄 Reconnect to load the model list, then pick a model.
    • Default is http://127.0.0.1:11434.
    • Custom host/port: change url to point anywhere — e.g. http://192.168.1.50:11434, http://my-server:11500, or even localhost:11434 (the scheme is added for you) — then click 🔄 Reconnect.
  4. Edit instruction to tell the model what to do (default: rewrite the prompt to be more vivid and detailed, output only the improved prompt).
  5. Wire the node's result output into your CLIP Text Encode.
  6. Tune keep_alive_minutes (minutes Ollama keeps the model loaded; -1 = forever, 0 = unload immediately) and timeout (max seconds to wait for a response). Toggle enabled off to pass prompts through unchanged.
  7. Leave trim_output on (default) to keep only the prompt — it strips any chat wrapper the model adds, e.g. a "Sure! Here's an improved version:" preamble, --- / ``` / Prompt: headers, surrounding quotes, and trailing offers like "Let me know if you'd like changes!". Turn it off to use the model's raw reply.

Outputs: result (the improved — and, by default, trimmed — prompt), context (Ollama context for chaining), meta (JSON with model, timings, token counts, and a trimmed flag). The result is also shown on the node.

Compatibility: the node uses the official ollama library with version-tolerant response handling (works across ollama-python 0.4→0.6 and recent Ollama servers), and declares VALIDATE_INPUTS so the dynamically-fetched model list passes ComfyUI's backend validation on current and older ComfyUI builds.

Outputs

| Output | Description | |--------|-------------| | positive | The positive prompt text for this row | | negative | The negative prompt text for this row | | filename_tag | Short tag you can append to saved filenames | | current_row | Row index (0-based) being processed | | total_rows | Total number of rows in the CSV | | is_last_row | True on the final row (handy for stopping downstream logic) |

Tips

  • Column names are flexible — you can name them anything. Set positive_column, negative_column, and tag_column to match your headers.
  • Reset: Use the Reset Counter node, or set reset_counter → yes on the loader node.
  • Manual mode: Set mode → manual-index and use manual_index to pick a specific row for testing.
  • The counter wraps around — after the last row it goes back to row 0 automatically.
  • The row state is saved in state.json inside the node folder, so it survives ComfyUI restarts.

Workflow tip: Save with filename_tag

Connect filename_tag output to a text node, then concatenate it with your save path in the Image Saver node so each image is named after its prompt row.

License

ComfyUI-BulkPrompt is licensed under the MIT License (see LICENSE).

The 🦙 Ollama Improver node is built on the official ollama Python library (MIT-licensed), installed separately via pip.