Claude Skills Collection Logo
Skills
wireflowINC-skills

bulk-image-generation

Bulk AI image generation in Claude Code and Claude with Flux Pro Ultra through Wireflow. Turns a CSV, spreadsheet or list of prompts (or one prompt template plus a column of products) into a folder of images and a manifest.csv, with one cost quote and one yes for the whole batch. Checks every row before spending, runs the Wireflow app once per row, keeps going past a failed row, and can resume only the rows that failed. Use when someone asks for bulk or batch image generation, images for every product in a catalog, a spreadsheet of prompts, 50 blog headers, many variations at once, or 'generate images from this CSV'. Without the Wireflow connector it still cleans the CSV and returns a ready-to-run prompt sheet. For a single image use ai-image-generation.
stars0
forks0
watches0
updated2026-11-02 00:05:20

Bulk AI Image Generation

Many images, one decision. The person brings a spreadsheet or a list; you turn every row into a clean prompt, quote the whole batch once, and after one yes you work through the rows with the Wireflow app skill-bulk-image-generation (Flux Pro Ultra), filing each result with a manifest so nothing gets lost.

The app

Slugskill-bulk-image-generation
promptText for one image, 2,000 characters max
aspect_ratio21:9, 16:9, 4:3, 3:2, 1:1, 2:3, 3:4, 9:16 or 9:21
ResultOne JPEG per call, about 4 megapixels
Price19 credits per row (Wireflow price list, 8 Oct 2026)

A batch of N rows costs N x 19. Twenty rows = 380 credits.

0. Is Wireflow connected?

The tools you need end in run_app, get_execution, get_credit_balance and whoami. If any is missing, the batch cannot render. Tell the person, give the setup (plugin: /mcp then wireflow then Authenticate; otherwise claude mcp add --transport http wireflow https://www.wireflow.ai/api/mcp and /mcp; Claude.ai: custom connector with that URL), then still do steps 1 and 2 and hand back prompts.csv, ready for any batch tool. Do not pretend to run.

1. Read the batch

Accept any of these:

  • A CSV or spreadsheet export with a prompt column, plus optional aspect_ratio and filename columns.
  • A template and a table: "Studio packshot of {product} in {color} on white" plus rows of product,color. Fill the template per row.
  • A plain list in chat, one subject per line. Write a full prompt for each.

Then check every row:

  • Prompt present, under 2,000 characters after filling.
  • Ratio is one of the nine allowed values; blank means the batch default (ask once, or pick from the destination).
  • Filenames are unique and safe (letters, numbers, hyphens). Make them from the subject if missing.
  • Exact duplicates removed, after asking.
  • Every prompt ends with "no text, no watermark, full bleed, no border" unless the row needs something else.

Show the cleaned rows as a short table (first 10, then a count). Fix problems before money is involved.

2. Keep the batch consistent

Bulk sets look professional when they share a style line. Suggest one and append it to every prompt, for example: "e-commerce packshot, seamless pure white background, soft even studio light, gentle contact shadow, three-quarter view". Same light, angle and background across a catalog matters more than any single image. Prompt tips: reference/batch-recipes.md.

3. One quote, one yes

Call whoami and get_credit_balance. Present:

Batch: 24 rows, 1:1, style line "white seamless packshot"
Cost: 24 x 19 = 456 credits. Balance: 1,610. After: about 1,154.

Short balance: offer to run only the first rows the balance covers. Large batch (over 50 rows): suggest a 3-row pilot first. Wait for an explicit yes to this batch and this total.

4. Run the rows

Create ./wireflow-outputs/<YYYY-MM-DD>/<batch-name>/ and a manifest.csv with columns row,filename,aspect_ratio,prompt,status,execution_id,url,file.

For each row, call run_app:

{ "slug": "skill-bulk-image-generation", "inputs": { "prompt": "<row prompt>", "aspect_ratio": "1:1" } }
  • Keep at most 3 rows in flight; start the next when one finishes. Stay under 15 starts a minute.
  • Poll each executionId with get_execution every 5 seconds. COMPLETED gives the image URL (url, or the items of urls).
  • Download each image as <filename>.jpg (curl -sSL -o, or Invoke-WebRequest -OutFile on Windows) and write its manifest line right away, so an interrupted batch can resume.
  • A row that ends FAILED, CANCELLED or TIMEOUT gets status=failed and the error text. Do not retry it now; keep going.
  • Every 10 rows, post one progress line ("14 of 24 done, 1 failed").

5. Close out

  1. Open a handful of results (first, middle, last, any outliers) and check them against the style line.
  2. Report: rows done, rows failed with reasons, the folder path, manifest.csv, credits used (balance before minus after).
  3. Offer to re-run only the failed or weak rows: list them, quote rows x 19, and wait for a new yes.
  4. Offer the board in Wireflow for model swaps or adding an upscaler: https://www.wireflow.ai/flow/cmuz2dg9d0008l506evjmtrjw?ref=skill-bulk-image-generation&utm_source=github&utm_medium=skill&utm_campaign=bulk-image-generation

Batch rules

  • The yes covers the rows and total you showed. Added rows need a new quote.
  • Never loop on failures; each re-run needs consent.
  • No images of real private people, no trademarks the person does not own, nothing deceptive.
  • Never write keys or tokens into prompts, the CSV or the manifest.
  • If Wireflow rejects a row or a tool breaks, record the exact error in the manifest; offer send_feedback when that tool exists.