Short answer / recommendation
Use a single Midjourney model/version and a single prompt template with fixed parameters (aspect ratio, seed, quality, no-text, background instruction) plus a reference-image for angle/lighting. Drive prompt text from a CSV of SKUs and product-copy, run prompts from a dedicated Discord server channel with a small automation script that posts prompts and downloads the resulting images, then rename files by SKU using the CSV mapping.

Why this works
Midjourney is stochastic by design; the only reliable way to get consistent angle/lighting/background across hundreds of SKUs is to lock every controllable variable: model/version, seed, aspect ratio, explicit lighting/camera wording, and (ideally) an image reference that demonstrates the exact pose/angle.

Prompt templates (use CSV variables like {SKU}, {NAME}, {REF_URL})
- Studio product shot (3/4 front-left), white seamless background, centered, natural softbox lighting, shallow depth of field, realistic photograph of {NAME} --ar 4:5 --v 6 --seed 12345 --q 2 --no text, watermark, logos
- Alternate (flat lay) Flat lay top-down photo of {NAME} on plain white background, consistent shadow on bottom-right, studio lighting, realistic, high detail --ar 1:1 --v 6 --seed 12345 --q 2 --no text
Tips: include one stable seed value (e.g., --seed 12345) and use the same --ar, --v and --q for all jobs. Use --no to exclude distractions. If you want a transparent PNG output, append --transparent (if your plan supports it) and remove background phrasing.

Upscaling & quality
- Two-stage: use Midjourney’s native upscale for a fast pass (the Upscale option in Discord), then run any final chosen images through a dedicated upscaler (Topaz Gigapixel, Real-ESRGAN) for print-level resolution.
- Keep original outputs and the upscaled versions; store metadata (prompt, seed, job id) in a CSV for traceability.

Automation hooks (practical, low-risk)
1) Dedicated Discord server: invite Midjourney bot and create a private channel. 2) Write a small script (Python + discord.py or Puppeteer) that reads your CSV of SKUs and product fields and posts formatted prompts (include REF_URL if you use a reference image). 3) The script watches the channel for the Midjourney bot’s reply message, downloads image attachments, and writes filename as {SKU}_{variant}.jpg. 4) Log JobID / message link into your CSV to link back to prompts.
Notes: if you have developer resources, build a proper Discord bot that uses a service account to avoid API rate limits. If an official Midjourney API is available to you, prefer that for scale.

Decision criteria (choose approach based on):
- Budget: manual + Discord is cheapest; paid API or dev time costs more but scales better.
- Skill level: non-coders can semi-manually paste prompts and use spreadsheet mapping; developers can fully automate.
- Throughput need: >100 images => automation strongly recommended.
- Output quality: for ecommerce, native MJ upscale + external upscaler gives best control.

Best-for / Avoid-if
- Best for: teams that need consistent, photorealistic SKU mockups quickly and can accept some post-QC.
- Avoid if: you require perfect photogrammetric accuracy (use real photography) or you need strict color/brand guarantees without manual color profiling.

Practical checklist
- [ ] Pick 1 model/version and lock it for the whole batch (--v 5/6 etc.)
- [ ] Create 1 prompt template and test with 5 representative SKUs
- [ ] Use a single seed, aspect ratio, quality for all runs
- [ ] Provide a reference image URL to lock angle/scale when possible
- [ ] Run small pilot, review 10–20 images, tune wording
- [ ] Build automation script to post prompts & download attachments, map filenames to SKUs
- [ ] Upscale final images (MJ native then optional external)
- [ ] Store prompt, seed, job ID, and final filename in a master CSV

If you want, I can give: (A) a ready-to-run prompt template filled with variables for your CSV, or (B) a short Python pseudocode script for posting prompts + capturing image attachments in Discord. Also note copyright/commercial-use settings and subscription limits when batching.

Compare Midjourney and Leonardo AI