Short answer / recommendation
Use a single Midjourney model version + one locked seed + fixed camera wording + low randomness and a reference image. Produce a “master” render that you lock to as the camera/lighting source and then swap only the product details (or use image-to-image with that master as the background). This yields consistent angle, scale and lighting across 30 catalog images.

Why this works (decision criteria)
- Determinism: --seed reduces composition randomness for a given model version. Always use the same version (v5/5.2/v6) because seeds behave per model.
- Prompt stability: identical camera/building-language and the same aspect ratio/parameter set produces consistent framing and scale.
- Reference anchoring: using one reference render or a photo as an image prompt gives MJ a visual anchor for angle, lighting and scale.
- Chaos/stylize: lower chaos and stylize = less creative drift.

Concrete prompt & parameter strategy (practical)
1) Pick a model version and lock it: e.g., --v 5.2 (or whatever you tested).
2) Choose an aspect ratio used across catalog (e.g., --ar 4:5 for vertical).
3) Produce a master reference: craft a full prompt that describes camera and lighting precisely: e.g. “product on neutral seamless, camera 30° elevation, three-quarter (45°) view, 85mm lens, 1.2m distance, 3-point studio lighting (softbox key upper-left), shallow depth of field, photorealistic”. Add any background instructions like “plain white background” or “shadow on floor” so the background stays identical.
4) Create the master image with a chosen seed: e.g., --seed 123456 --chaos 0 --stylize 50 --quality 2. Upscale and save the master.
5) For each product, either:
- Use the master as an image prompt: upload master.jpg and call it first in the prompt (master.jpg::0.8) then append the product description, or
- Use image-to-image (i2i) with master as base and a higher image weight so MJ preserves camera/lighting while swapping product details. Use same --seed and low --chaos.
6) Only change the product descriptor (material, color, label) — keep camera/lens/lighting text, aspect ratio, seed and parameters identical.

Practical checklist before batch generation
- [ ] Choose and note MJ model version.
- [ ] Choose and fix --seed (record it).
- [ ] Fix --ar, --v, --chaos (0–10), --stylize (low for photoreal), --quality.
- [ ] Make a master render and save as reference.
- [ ] Use master.jpg as image prompt or i2i base for all variants.
- [ ] Produce test batch (3–5 variants) and verify scale/angle.
- [ ] If any drift, tweak image weight or retry with different seed, then lock.

Best-for / Avoid-if
- Best for: stylized or photoreal ecommerce catalogs where exact camera identity (angle, lighting) is visually consistent but not dimensionally critical. Good when budget/time prevents full 3D workflow.
- Avoid if: you need CAD-accurate measurements, exact scale relations between items, or legally precise product photos — for those, use real photography or 3D renders.

When this changes based on team/budget
- Small team/low budget: use Midjourney + this anchored master approach.
- Large team/high fidelity needs: consider full 3D (consistent camera metadata) or a dedicated studio photographer.
- If you want help turning your product list into templated prompts, use a prompt-engineering helper (e.g., ChatGPT) to generate consistent prompt variants and CSVs for batch work.

Final tip
Keep a single “prompt spec” file (model version, seed, camera line, lighting line, aspect ratio, negative prompts) and enforce it for every render. That small discipline will produce the consistent catalog you need.

Compare Midjourney and Leonardo AI