Imagery at the speed
of your retail calendar.

No — and the proposal is structured specifically to make sure it doesn't. The pipeline only works if someone inside LiveTrends with deep product knowledge operates it. Generic AI without that domain expertise produces generic AI slop. With your photographer directing it, the output is on-brand at a scale you can't get any other way.

Phase 2 includes training your photographer on the workflow. By Phase 3 they're operating it solo. Their job becomes less about being the bottleneck on every shot and more about being the brand-fidelity gatekeeper across hundreds of generated assets per week.

Firefly Custom Models is the closest enterprise alternative. The honest tradeoff:

Firefly pros: integrated with Creative Cloud, no infra to manage.
Firefly cons: cloud-only (your designs sit on Adobe's servers), per-image pricing scales painfully at your volume, no operator embedding — Adobe doesn't send a person who knows your brand.

If you're already paying for Creative Cloud Enterprise, we can complement Firefly: train the on-brand LoRA here, hand it off to your Firefly workflow. The custom-training step is the differentiator either way.

Walmart and Target have published guidance requiring AI disclosure for some asset categories. We label generation metadata at the file level so compliance is automatic. Lowe's currently has no disclosure rule for marketing imagery. Most lifestyle assets are fine; we flag any that aren't.

Worth noting: many of your competitors are already using uncredited AI imagery — without the audit trail. The local-pipeline approach is actually more defensible in compliance reviews.

It's a LoRA (Low-Rank Adaptation) fine-tune on top of an open-weights base model — Stable Diffusion XL or Flux. Trained on your catalog. The model file is small (~50–200 MB), runs on a single workstation GPU (RTX 4090 / 5070 Ti / 5080-class). The pipeline orchestrator is ComfyUI, an open-source node-based workflow tool.

Nothing proprietary on the inference side. If you ever decide to take it in-house entirely, you keep the model. No vendor lock-in.

Phase 2 kickoff: roughly two weeks from a signed NDA. Breakdown: 1 week for catalog ingestion + initial training, 2–3 weeks for production generation + curation, final week for delivery + photographer training. Most efficient if scoped to one specific campaign with a real deadline (Lowe's spring reset, holiday Target endcap, etc.) — gives the work a measurable comparison point.

Two paths:

Replace: If the agency is mostly producing variants you could generate yourself with the right tooling, the embed pays back in months. Most agencies charge $200–800 per finished image; we land closer to $40–80 marginal cost at the volumes you'd run.

Complement: If the agency does work that needs human creative direction — campaign concept, photo-real packaging shots, brand films — keep them for that. Use the trained pipeline for the lifestyle-variant tail (the 80% that's currently uneconomic to commission).

Honest answer: a bigger agency would charge $200–500K to do this and the deliverable would be a slide deck and a SaaS subscription. We're solo (Orlando, FL), local, and the cost structure reflects that. Phase 2 is $7.5–12K for a real working pipeline.

The other answer: this proposal only works if someone inside LiveTrends operates it. A bigger agency would never structure the deal that way — they need you dependent on them. We're structuring it the opposite direction: by Phase 3, you don't need us for day-to-day. We stay on for the model-engineering work that requires the specialty.

One workstation with an RTX 4090 / 5070 Ti / 5080-class GPU (16–24 GB VRAM). ~$3K hardware, one-time. Optional: a second machine for batch runs if you want to generate overnight without tying up the operator's box. Everything else runs on standard Windows or Linux.

If you don't want to provision hardware, we can run the pipeline on our own infrastructure during Phase 2 and revisit hardware at Phase 3.