Workflow: Automate Midjourney batch mockups for 50 SKUs

Recommendation (short):
Use a reproducible prompt template + a CSV-driven job list, generate the 50 prompts programmatically (ChatGPT or a script), feed them to Midjourney in controlled batches, keep a consistent seed and model/flags, then upscale with a reliable CLI/desktop upscaler and apply a scripted filename/mapping and Photoshop/Imagemagick mockup overlay. Keep a human-in-the-loop for final QA.

Why this approach:
- Prompts + seeds = repeatable outputs across many SKUs.
- Batch-driven automation (CSV → prompt → send → fetch) minimizes manual typing.
- Post-upscale and template compositing ensure uniform mockup frames.

Decision criteria (pick path based on constraints):
- Budget: if you can pay for Topaz Gigapixel and a Midjourney Pro plan, go fully automated + high-quality upscaling. If not, use Real-ESRGAN and the relaxed MJ plan.
- Skill: if you have dev skills, automate Discord interactions with Playwright/Puppeteer (note: check TOS). If not, use semi-manual paste + CSV and a clipboard tool.
- Output quality/brand compliance: strict product photography needs real shoots; MJ is best for stylized photoreal mockups.

Prompt template (use variables: SKU, color, material, view, seed):
"Photoreal product mockup of [SKU] in [COLOR], on seamless white background, studio softbox lighting, 45° angled view, realistic material texture, high-detail, 50mm lens, shallow depth of field, no props --ar 1:1 --v 5 --q 2 --stylize 30 --seed [SEED]"
Example for one SKU:
"Photoreal product mockup of 'Acme Mug - SKU123' in matte black, on seamless white background, studio softbox lighting, 45° angled view, realistic ceramic texture, high-detail, 50mm lens, shallow depth of field, no props --ar 1:1 --v 5 --q 2 --stylize 30 --seed 11111"
Tips: keep --ar consistent (1:1 for ecommerce thumbnails), set --seed per SKU for reproducibility, use --no to remove unwanted elements (e.g., --no text). Use an image reference (upload a clean PNG silhouette) if you need exact shape consistency.

Upscaler recommendations:
- Production/high quality: Topaz Gigapixel (best detail, paid).
- Free/CLI friendly: Real-ESRGAN or waifu2x-caffe for bulk processing.
- Midjourney internal: use --uplight or --upbeta for faster/lightweight upscales; still do an external upscale for final assets.

Filename convention (keep machine-friendly and traceable):
SKU_Color_View_Seed_Model_Version.png
Example: SKU123_MatteBlack_45deg_seed11111_v1.png
Keep a CSV: columns = SKU, color, prompt, seed, MJ_job_id, final_filename, QC_status.

Automation approach (two options):
1) Semi-automated, recommended for most teams: generate prompts via ChatGPT or a script; copy/paste into Discord in small batches (4–8 at a time), monitor, upscale, download, run batch upscaler, run compositing script to place into mockup template.
2) Full automation (advanced): use Playwright/Puppeteer to control a logged-in browser to post MJ prompts and retrieve images programmatically. WARNING: verify Midjourney and Discord terms of use before running headless bots; maintain rate-limits and a dedicated account.

Practical checklist (ready-to-run):
- [ ] Prepare CSV with 50 SKUs and variables.
- [ ] Create prompt template and a seed-per-SKU plan.
- [ ] Optionally create a reference silhouette PNG for shape consistency.
- [ ] Generate prompt list (ChatGPT or script).
- [ ] Choose automation route (semi-auto vs full).
- [ ] Queue/send prompts in controlled batches to Midjourney.
- [ ] Upscale chosen outputs (internal then external if needed).
- [ ] Rename and map files per CSV.
- [ ] Composite into mockup frames and run QC.

Best-for / Avoid-if:
- Best for: fast, consistent photoreal stylized mockups at scale.
- Avoid if: you require pixel-perfect brand photography, or if you cannot accept potential model variance across outputs.

Tools to use: Midjourney for generation and ChatGPT to help generate and parameterize prompts and CSVs.

If you want, I can: generate the 50 prompt rows from your SKU list (need CSV) or produce a Playwright pseudocode snippet for posting prompts to Discord (with TOS warnings).

Compare Midjourney and Leonardo AI