Short answer
Use a single “studio master” prompt + fixed seed and a reproducible img2img / inpaint pipeline so every SKU is generated into the same camera/lighting frame. Use reference images and a mask to keep placement, then only swap the product visual data.

Recommendation
Create one master scene (empty studio or placeholder product) that encodes exact camera, lens, lighting, background and floor reflections. Save that as your template reference and use it as the base image for all SKU generations with the same seed and prompt template. This gives the most consistent results for a 150-item catalog.

Why this works (decision criteria)
- Use pure text-to-image when you have no photographic base and want speed, but consistency will be harder. Use img2img/inpaint when you can supply a template or a photographed product cutout; it preserves geometry and lighting.
- Fix the seed, sampler, resolution, and CFG/strength to lock randomness. If you must vary style or realism, expect more manual QC.
- Choose methods based on budget/skill: automated img2img batches are cheaper in time and scale-friendly; hand-touch retouching yields higher quality but costs more per SKU.

Practical checklist (step-by-step)
1) Create a master template
- Make a neutral “empty studio” render: same canvas size/crop you need (e.g., 2000×2000), neutral white/grey seamless background, consistent shadow plane. Save as reference.
2) Build an explicit master prompt (example below) and negative prompt
- Include precise camera & lens details, light positions, background, and shadow behavior. Add negative prompts to avoid variations.
3) Lock generation settings
- Fixed seed (e.g., 12345678), sampler (Euler a / DPM++ whatever you prefer), steps (30–50), cfg (6–9), same resolution.
4) Prepare product assets
- If you have clean PNG cutouts or raw studio photos, place them into the template and mask the product area. If not, use the product image as the img2img input with low strength (0.3–0.5).
5) Use img2img/inpaint with the template
- Keep mask shape/position identical for every SKU so the camera angle and light falloff remain the same.
6) Batch generation and QC
- Run 10–20 test SKUs, check consistency (angle, shadow, specular), then process in batches. Flag items that need manual retouch.
7) Final passes
- Simple color-correct and crop to the same bounding box. Save metadata for traceability.

Sample master prompt (adapt to Leonardo AI wording)
"Studio product shot, 50mm lens, f/5.6, ISO100, 1/200s, camera height 45 cm, 3/4 frontal view (approx 15° turn), softbox key light 45° top-left, fill light 70% opposite, subtle back rim light, white seamless background, soft shadow on right, realistic material reflections, high detail, photorealistic"
Negative prompt: "text, watermark, extra objects, floating parts, extreme perspective, distorted anatomy"

Seed strategy
- Use one fixed seed for the whole catalog if you want near-identical framing. If you need slight variation, use a small set of seeds (3–5) per angle and rotate.
- If switching products changes silhouette a lot, use img2img with low strength to preserve lighting/angle but let the model adjust shape.

Best-for / Avoid-if
- Best-for: large catalogs needing identical camera angle and lighting for comparison (e-commerce). Works well if you can supply cutouts or a base template.
- Avoid-if: you need wildly different artistic styles per SKU, or if products require custom staging that can't be templated.

Notes on resources
- More automation reduces per-item labor but requires spending time to build a robust template and mask system. Small teams can get good results with img2img; larger teams should add a quick manual QC step.

If you use Leonardo AI, save your template, seed and prompt as a preset and run batches from that preset for maximal reproducibility.

Compare Leonardo AI and Midjourney