Recommendation:
Build a small set of “master reference” images (1–5) that contain the exact lighting, camera angle, and pose you want, then reuse those as image prompts + a locked parameter set (seed, aspect ratio, stylize, model version). Generate garments by swapping only the garment text tokens (or by using inpainting/masks) so lighting/perspective remain consistent.

Why this works (short):
Midjourney’s seed + fixed parameters make composition more repeatable, but the easiest, most reliable way to keep lighting/perspective/pose identical across hundreds of outputs is to anchor the generation to real reference images and minimize creative variation (--stylize low, same model version).

Decision criteria (pick approach by constraints):
- Low budget / smaller batch (tens): manual reference + consistent prompt/seed works fine.
- High volume (hundreds+): create 2–3 master reference images and use inpainting or image-pair prompting; automate prompt submission via a scripted Discord bot or a batch workflow.
- Skill level: inpainting and script automation require intermediate technical skill. Non-technical teams should standardize templates and hand off to one operator.
- Output quality vs speed: lowering stylize and using seed gives consistency but can make images less “creative”; inpainting keeps quality high with tight consistency.

Concrete prompt strategy and parameter checklist (use as template):
1) Create master reference(s): photograph or render a model in your chosen pose/lighting on neutral background. Upload to Discord and copy image URL.
2) Use a fixed prompt template (replace GARMENT_TEXT):
IMAGE_URL "garment: GARMENT_TEXT" studio softbox lighting, single key light 45 degrees from camera, neutral gray seamless backdrop, model facing camera 3/4 view, hands relaxed at hips, natural pose, 50mm, f8, shallow but consistent DOF --ar 3:4 --v 5 --s 50 --q 2 --seed 123456
Notes: adjust --v (model version) to your account’s recommended stable version; lower --s (stylize) to reduce artistic drift; pick one --seed and reuse.
3) If using image-weighting, append (for MJ v5+): --iw 1 to force the reference to dominate.
4) Use --no text, --no logo to remove unwanted elements.

Best-for / Avoid-if
- Best for: product catalogs, consistent apparel mockups, A/B comparison, ecommerce images.
- Avoid if: you need wildly different artistic interpretations each image (increase --s then), or if you can’t supply a reliable master reference.

Practical checklist to run a batch reliably:
- [ ] Shoot / create 1–3 master reference images for each pose/angle you need.
- [ ] Pick and fix parameters: --ar, --v, --s (low), --q, and one --seed per reference.
- [ ] Build a prompt template and a CSV of garment variables (color, pattern, texture, label).
- [ ] Test 5 variants per master to confirm consistency; tweak --s and --iw if needed.
- [ ] For garment-only swaps use inpainting/mask workflow so body/pose/lighting are untouched.
- [ ] Automate submissions (scripted Discord bot or manual copy/paste if small scale).
- [ ] Track which seed + reference produced each output for reproducibility.

Tips & caveats:
- Seeds are not perfectly deterministic across major Midjourney model updates; always specify --v and retest when the model version changes.
- Image prompting + inpainting is the most reliable to preserve pose/lighting; pure text prompts are more variable.
- If you want help generating many prompt variations from a template (CSV -> prompts), use ChatGPT to expand and format the text batch quickly.

If you want, I can: 1) give a ready-to-run prompt CSV template, or 2) craft the exact prompt you should paste into Discord for MJ with your preferred aspect and pose. (I recommend Midjourney as the generation tool.)

Compare Midjourney and Leonardo AI