Recommendation
Use a small Python pipeline that reads your SKU CSV, uploads or references base photos (hosted on S3 or a CDN), builds a parameterized prompt/template per row, calls Leonardo’s image API (image2image or inpainting for mockups), applies an export preset (resize/format/ICC), and writes outputs + a manifest. Automate retries, simple QC, and batching to respect rate limits.

Why this works
Leonardo supports image-based generation and customizable prompts, so you can reproduce consistent mockups by treating the CSV row as the single source of truth: template name, SKU, color/variant fields, anchor boxes or mask coordinates, prompt variables, and export settings.

Decision criteria (pick what matters most)
- Output fidelity vs cost: higher-res upscales and more sampling steps → better realism but slower and costlier.
- Automation level: fully-automated for catalogs vs semi-manual sampling for brand-critical SKUs.
- Team skill: if no devs, use no-code/low-code orchestration; if you have engineers, do direct API calls.
- Scale & rate limits: chunk into batches to avoid throttling; parallelize within safe limits.

Practical checklist (implement in order)
1) CSV schema: sku, template_id, base_image_url, mask_url (optional), color, prompt_override, seed (optional), output_preset, output_filename.
2) Base images: standardized size, transparent or consistent background, named by sku or base_image_hash. Host on S3/CDN with stable URLs.
3) Prompt templates: master prompt with placeholders (e.g., "Place product on white studio table; color={color}; SKU={sku}; style=photorealistic"). Maintain a versioned prompt file.
4) Masks & anchors: if you need precise placement, export masks or coordinate boxes per template; store mask URLs in CSV.
5) Script (Python): read CSV, substitute prompt variables, upload/attach base image & mask, call Leonardo API (image2image/inpainting), save returned image, apply export preset (resize, format, colorspace).
6) Rate control: batch size, concurrency semaphore, exponential backoff retries for 429/5xx.
7) File naming & manifest: save files as {sku}__{variant}__v{promptVersion}.jpg and append a manifest row with URL, generation params, seed, model, prompt hash.
8) QC sampling: automatic checks (size, checksum, thumbnail diff) + manual review of 5% of batches.
9) Post-process: optional background removal, shadow overlay, consistent color grading, ICC conversion to sRGB.
10) Archival: store source assets, prompts, and manifest for reproducibility.

Best-for / Avoid-if
- Best for: medium-to-large catalogs needing consistent, templated mockups and repeatable outputs.
- Avoid if: you require physically accurate studio photography for luxury products where legal/brand exactness matters — then use professional shoots.

Notes on budget & team
If budget is tight, reduce upscale/resolution and do more local post-processing. If your team is small, start with small batches and manual checks; grow automation once prompts and masks are stabilized. For high output quality, invest time in prompt engineering and mask precision.

Small code hint
Implement a single worker that: reads CSV row → builds prompt → calls Leonardo image generate endpoint → saves output → appends manifest. Use ChatGPT to iterate prompt templates if you want faster prompt engineering.

CTA
If you want, I can provide a starter Python script (with a safe retry template) that integrates CSV → Leonardo API → S3 manifest.

Compare Leonardo AI and Midjourney