Wujiai 通用文件夹批量处理 (CSV/Excel)

Most batch nodes read a spreadsheet and call it a day. This one reads a spreadsheet and a folder of images, zips them together, and runs image-to-image style generations across the whole set - the "scan a local folder of images, combine with CSV prompts, concurrent batch img2img, auto-save to a target directory" feature from the pack README. If you have 300 reference photos and a CSV of 300 prompts and you want a consistent edit applied to all of them, this is the node. It's the heavyweight of the Wujiai Universal family.

Inputs:

• model_name - any Wujiai image model, default gemini-3.1-flash-image-preview.
• directory_path - the folder to scan for images.
• csv_file_path - the CSV/XLSX with per-file prompts.
• column_name - the prompt column (default prompt).
• aspect_ratio, image_size, sequential_mode, max_images, n - the usual generation knobs, applied per job.
• executions_per_prompt - variations per row.
• max_concurrent_files - 1–100, how many folder images are processed in parallel (default 3). This is the speed-versus-rate-limit dial.
• fixed_prompt - optional; overrides the per-row prompts with one fixed prompt applied to every image.
• api_key - optional, pack-wide fallback.

Outputs:

• images_preview_15 - a preview IMAGE batch (capped around 15 images) so you can eyeball what came out without hunting through folders.
• status_report - STRING with per-file results, counts, and the usual credit balance.

Mechanically it matches folder images to CSV rows (by order or index), launches concurrent image-to-image jobs through the Wujiai gateway, and saves results to the target directory with the filename/ordering from your inputs. The max_concurrent_files throttle matters more here than anywhere else in the pack: a folder of hundreds of images is exactly the load that gets you rate-limited or auto-banned on a paid gateway, and 3 is a sane default for that reason.

Install is pack-wide:

cd ComfyUI/custom_nodes
git clone https://github.com/jieg9341-lab/ComfyUI-nkxx

Restart; auto-install covers pandas, openpyxl, requests, yt-dlp, opencv-python, aiohttp. Register at wujiai.org for the key.

Grounded gotchas:

• Folder size × cost. This node will happily process hundreds of paid generations. Budget before you run, not after.
• Prompt/image pairing is positional - if your CSV rows and folder files are out of order, you get edits applied to the wrong subjects. Sort both before pointing the node at them.
• The preview batch is a convenience, not the deliverable; the real outputs are written to disk. Don't build downstream nodes off the preview and expect all 300 files there.
• max_concurrent_files is your rate-limit insurance. Crank it to 100 only if you're sure the gateway can take it; start at the default and watch the status_report for errors.