Short answer / recommendation
To force consistent photoreal product shots in Midjourney, lock the model/version + seed, use low stylization, minimize chaos, prefer a reference photo or low-denoising img2img workflow, and use the Light/upscaler that preserves real detail. Those five controls usually turn painterly results into repeatable, photographic renders.

Why this happens (brief)
Midjourney balances artistic “style” and randomness with your camera/lighting tokens. If stylize, chaos, upscaler, model version, or seed change between runs you’ll get painterly variations even if the rest of the prompt is identical. Also starting from pure text (T2I) often drifts stylistically; seeding or using an image anchor fixes the look.

Decision criteria (which knob to turn)
- Need absolute realism and repeatability (product catalog): use a photo reference + img2img with low denoise, fixed seed, --stylize near 0, low --chaos, and Light upscaler. Paid plan recommended for higher quality runs and faster throughput.
- Starting from concept art but want realistic feel: use text-to-image but set --stylize low and run with a fixed seed; expect more revisions.
- Fast iterations and many variants: accept slightly higher stylize (10–50) and use multiple seeds, then pick best and refine with img2img.

Practical checklist (copy/paste each job)
1) Lock model/version: start every run with the same MJ version (e.g., /settings -> choose the same model).
2) Set seed: append --seed to reproduce exact framing/details.
3) Reduce stylization: append --stylize 0 (or 0–50) so MJ prioritizes your literal prompt over painterly flourishes.
4) Reduce randomness: append --chaos 0–10.
5) Increase fidelity if budget allows: append --quality 1 or --quality 2 for better detail (higher cost/time).
6) Use negative terms: add “no painting, no brushstrokes, no watercolor, no illustration” to the prompt.
7) Use a reference image: upload a clean product photo and use it as a prompt anchor; for changes, switch to img2img with denoise ~0.2–0.4 to preserve realism.
8) Upscaling: use the Light/Uplight upscaler or the “preserve detail” option to avoid adding painterly detail; avoid aggressive upscalers that invent texture.
9) Be explicit about camera: e.g., “studio product shot, full-spectrum softbox, 50mm, f/8, 1/125s, natural color, white seamless background, true-to-life materials.”
10) Batch with consistency: keep the same aspect ratio (--ar) and framing for all images in a product set.

Best-for / Avoid-if
- Best-for: product catalogs, consistent e-commerce imagery, small variations on a photographed product, teams that need reproducible assets.
- Avoid-if: you want creative, stylized hero shots or painterly editorial images — then higher stylize and more chaos are desirable.

Extra tips
- If you have a real sample, invest the time to photograph a neutral reference and use it for img2img—this gives the highest consistency.
- Test one configuration per product: run a few seeds and pick the seed you like, then reuse it for all angles.

When to change approach
If budget or time is limited, accept slight stylization and do manual cleanup in Photoshop. If you need pixel-perfect brand consistency across many SKUs and have a team, use a tightly controlled pipeline: fixed prompts, seeds, img2img references, and a checklist above.

If you want, I can review a sample prompt (and the model version you’re using) and give exact parameters to paste into Midjourney.

Compare Midjourney and Leonardo AI