Recommendation (short):
Use one master “angle proof” render with a fixed seed and camera-language in the prompt, save that image, then drive all color/finish variants by using that master image as an image-prompt + the same seed and identical generation settings (--v, --ar, --stylize, --q). This gives highly consistent angle/perspective across variants. For absolute precision (pixel-perfect alignment) use a 3D render pipeline instead.

Why this works (brief):
- Midjourney’s --seed fixes the noise starting point; using the same seed + identical model/version and prompt template keeps composition/perspective much more consistent.
- Feeding the master render back as an image prompt anchors the model to the exact framing and lighting, so subsequent prompts mainly change surface details (color, texture) instead of viewpoint.

Decision criteria (pick strategy):
- Need near-perfect alignment for 1:1 product galleries and have 3D skills/budget? Use a real 3D render pipeline (best consistency).
- Want fast, low-cost AI-native variants? Use the Midjourney master-image + same-seed strategy below.
- Team size/skill: if you have a single operator, the image-prompt workflow is simplest; larger teams should store the exact prompt + seed in a registry to ensure reproducibility.

Practical step-by-step checklist (use exactly in this order):
1) Choose constants: model version (e.g., --v 5), aspect ratio (--ar 4:5 or 1:1), quality (--q 2), stylize low (e.g., --stylize 50 or lower for realism). Record them.
2) Write a camera/focus line in the prompt: e.g., “product shot, 3/4 view front-left, camera at product center height, 45° pitch, 50mm lens, shallow depth of field f/5.6, studio key+fill lighting, white seamless background.” Keep it precise and reuse verbatim.
3) Create a neutral master prompt that describes the product shape/materials neutrally (no color-specific terms). Add a fixed seed: --seed 123456 (pick any integer). Generate 4 variations and pick the best.
4) Upscale and download the chosen master image at the highest quality. This is your anchor reference.
5) For each variant, prompt using the master image URL at the start, keep the same exact prompt text (camera line unchanged), then add only the surface change instruction e.g., “replace material with matte navy leather” or “color: Pantone 286C, gloss 30%”. Include the same --seed and all the same parameters.
6) If small artifacts shift the shape, increase image-weight (if available) or explicitly add phrases like “exact same pose and perspective as reference image.” Repeat until consistent.
7) Run a visual QA: overlay images in Photoshop to check alignment. If unacceptable, switch to a 3D pipeline.

Best-for / Avoid-if
- Best for: catalog-style e-commerce images where you need the same shot composition across colors/finishes and you don’t require pixel-perfect registration. Works great for small catalogs, rapid iteration, low budget.
- Avoid if: you need absolute, pixel-perfect alignment, exact measurements, or automated bulk rendering at scale—use 3D/photography instead.

Notes and caveats
- Midjourney’s --seed reduces randomness but does not guarantee perfect determinism across model updates. Lock the model version (--v) and save the seed & full prompt.
- If you want automation (bulk variants), script calls to Midjourney via the bot or use a saved prompt registry and QA step.

If you want, I can draft a copy-paste prompt template (with camera line and example seed) tuned to your product category. For hands-on image prompting, use Midjourney to run the workflow and store the master render.

Compare Midjourney and Leonardo AI