Need Help, hiring an AI Automation Architect: Production-Grade "Content Factory" for E-commerce

Hire a Pro Hire Help
1

I am seeking a high-level AI Automation Architect to build a fully automated (or semi-automated) content creation pipeline for a home decor brand specializing in Table and Floor Lamps. The goal is to bypass manual content creation by using an orchestrator to generate 100% AI-driven lifestyle assets.

The Core Challenge: Most AI workflows “hallucinate” or warp product shapes. This project requires a “Preservation Strategy” where the lamp is treated as an immutable foreground layer—its pixels are never re-generated. Additionally, the system must master lighting physics, ensuring the lamp acts as the primary light source with realistic shadows, glow, and light spill.

Key Responsibilities & Workflow Design: You will be responsible for building a Make.com orchestration layer that connects the following stages:

1. Intake & Segmentation: Triggered by a photo upload, the system must use Photoroom API or SAM 2 to generate high-res transparent cutouts and binary masks.

2. Scene Generation (The “Room Plate”): Using ComfyUI (SDXL or FLUX), the system must generate a “cozy, moody, cinematic” background room while leaving an empty space for the product.

3. Lighting Integration Pass: A crucial two-pass process. After compositing the lamp, the system must run a masked inpainting pass (using ControlNet/IP-Adapter) to generate realistic light spill and contact shadows around the product without touching the product itself.

4. Short-Form Video Production: Automate the creation of 9:16 vertical clips (Reels/TikTok) using Runway Gen-3 Alpha Turbo or Creatomate. The strategy must focus on animating the background only or using subtle camera movement to maintain 100% product integrity.

5. Automated QA & Guardrails: Implement a final vision-based check (using CLIP or ViT-style APIs) to compare the final asset against the original photo, flagging any warped or distorted products for review.

6. Copywriting & Scheduling: Integrate ChatGPT API for SEO-optimized captions and hooks, then push final assets to Buffer or Later.com.

Required Technical Expertise:

• Make.com Orchestration: Expert-level ability to connect complex media pipelines via Webhooks and APIs.

• ComfyUI Mastery: Deep knowledge of node-based workflows, ControlNet (Canny/Depth), and IP-Adapters for style and geometry locking.

• Lighting Physics: Ability to engineer prompts using technical terms like “Subsurface Scattering,” “Chiaroscuro,” and “Volumetric Fog” to ensure photorealistic light behavior.

• API Integration: Experience with Photoroom, Runway, Creatomate, and OpenAI.

Aesthetic Goal: The final output must consistently reflect a “cozy, moody, cinematic” atmosphere with high-end, magazine-quality lighting.

Deliverables:

• A functional Make.com blueprint.

• A ComfyUI node graph optimized for product preservation.

• A structured Airtable or Google Sheets “Source of Truth” database

5 Likes
2

Hello @Jefferson_Pak , welcome to make.com community, I have worked and have experience with Make.com and l will love to collaborate with you on this you can schedule a call Here and you can checkout my upwork profile Here, for my pastworks and certifications

1 Like
3

Hey @Jefferson_Pak , I see you’re trying to solve the hardest part of AI product content: preserving the lamp 1:1 while still achieving cinematic, physically believable lighting at scale. I can design a Make.com–orchestrated pipeline where the product remains an immutable foreground layer, with ComfyUI handling room plates, masked lighting passes, and background-only motion for video.

I’ve built similar multi-stage AI systems using segmentation + ControlNet/IP-Adapter guardrails, vision-based QA checks, and API-driven orchestration. I’m confident I can deliver a stable, repeatable pipeline that produces on-brand, magazine-quality assets without product distortion.

Schedule a call here to discuss this.

4

This project makes a lot of sense, and I understand exactly what you’re trying to solve, especially the issue of AI warping products and breaking lighting realism.

I’m an AI Automation Architect with strong experience in Make.com, and I build production-ready automation pipelines, not experimental AI workflows. My focus is always on preserving the original product while letting AI handle everything around it.

How I would handle this

Product preservation

  • The lamp is treated as a fixed foreground asset

  • High-quality cutouts and masks created using Photoroom API or SAM 2

  • The product itself is never regenerated or edited by diffusion models

Make.com orchestration (my core strength)

  • Webhook-based intake when images are uploaded

  • Clean, modular Make scenarios that are easy to scale and maintain

  • Clear flow across segmentation, background generation, lighting pass, video, QA, and scheduling

  • All assets and metadata saved to a single source of truth (Airtable or Google Sheets)

ComfyUI setup

  • Background “room plate” generated with SDXL or FLUX

  • Depth and Canny ControlNet for realistic room structure

  • Empty space reserved for the lamp

  • Separate masked inpainting pass to add:

    • Light spill

    • Contact shadows

    • Glow and ambience
      without touching the product itself

Short-form video

  • Background-only animation or subtle camera movement

  • Product remains completely static

  • Automated using Runway Gen-3 or Creatomate

Quality checks

  • Vision-based comparison (CLIP / ViT style)

  • Flags any output where the product shape or proportions change

Copy & scheduling

  • AI-generated captions and hooks

  • Automated posting via Buffer or Later

What I can deliver

  • A working Make.com blueprint

  • A ComfyUI node graph built specifically for product safety

  • An Airtable or Google Sheets control system

  • Clear documentation so the system is easy to run and extend

If this sounds like what you’re looking for, message me for my portfolio, pricing, and more technical details.

Email: fopefoluwaakinola@gmail.com

Happy to walk you through the full setup and adjust it to your brand’s exact look and scale.

5

This is very aligned with the kind of production-grade content automation systems we build at Hashlogics. We’ve worked extensively on AI pipelines where product preservation is non-negotiable, and the orchestration layer (Make/n8n) is responsible for enforcing guardrails rather than just “generating images.”

A few relevant references to how we think about this:

• ESG Sustainability Platform (Greenlight → Studio Birthplace) – shows how we design deterministic, auditable pipelines with strict validation and QA layers rather than free-form generation
Case study: https://drive.google.com/file/d/1Ace8kGEDlAXbScKwezalrV5aSESY4B9p/view

• QA.Tax – multi-stage automation with structured outputs, validation passes, and failure flagging (similar philosophy to your CLIP/ViT QA step)
Loom walkthrough: https://www.loom.com/share/e65b8f51ca0c426bac85751817cebb3c

• SearchAtlas AI Pipeline – large-scale orchestration where assets are generated, checked, routed, and only published if they pass quality gates
Loom walkthrough: https://www.loom.com/share/e72ea66f588841078dcf961e6a69d077

On your architecture specifically, the immutable foreground + two-pass lighting integration approach is the correct direction. Treating the lamp as a locked alpha layer, then using masked inpainting purely for light spill, contact shadows, and bloom is how you avoid geometry drift. Pairing that with background-only motion for short-form video is exactly how we’ve kept product integrity intact in similar systems.

Happy to walk you through how we’d structure the Make blueprint, ComfyUI graph, and automated QA gates end-to-end. If useful, let’s do a short call and I can show concrete orchestration patterns and failure handling:

:date: https://calendly.com/_hashlogics/abdul-basit

6

Jefferson

This is exactly my lane: I build production-grade AI content factories where the product remains a locked, immutable layer and only lighting, shadows, and environment are synthesized with physical accuracy.

If you want, let’s connect +234 904 684 2148 here and talk more about the project and start immediately

7

Hi Jefferson, welcome to the community.

This is a very well thought out brief, especially the emphasis on product preservation and lighting realism. Most content factories fail exactly where you’ve identified the problem.

I work as an AI Automation Architect focused on production grade Make.com orchestration for media heavy pipelines. I have experience designing workflows where the product remains an immutable foreground layer while all generative passes are constrained to masks and secondary layers.

At a high level, I would approach this by:

• Using a strict intake and masking phase where the product cutout and binary masks are treated as protected assets throughout the pipeline
• Generating room plates independently with spatial constraints so the product area is never touched
• Running a controlled lighting integration pass using masked inpainting only around the product boundary to simulate spill, contact shadows, and glow
• Handling video output by animating background layers or camera movement only, never regenerating the product
• Adding a final automated QA step to compare geometry and silhouette against the original product image before approval
• Orchestrating everything in Make with clear stages, retries, and guardrails, backed by a single source of truth in Airtable or Sheets

Happy to share relevant examples and discuss architecture privately if helpful.
You can either book a quick call with me or we can continue the discussion on Fiverr, whichever you prefer.

Best Regards,
Folafoluwa Olaneye.

8

Hi Jefferson,

This is exactly the kind of AI pipeline I build. The preservation problem you described is real, most workflows break product geometry and lighting.

I would design a Make.com orchestration where the lamp is treated as an immutable foreground layer. Segmentation via Photoroom or SAM 2, background only generation in ComfyUI (SDXL or FLUX), then a masked lighting pass using ControlNet or IP Adapter to add realistic light spill and contact shadows without touching the product pixels.

Video output would animate background or camera only using Runway or Creatomate, so product integrity stays 100 percent intact. A final CLIP or ViT check flags any warped assets before publishing. Captions and scheduling handled via OpenAI plus Buffer or Later, all tracked in Airtable or Sheets.

Deliverables include a working Make.com blueprint, a product safe ComfyUI graph, and a clean source of truth database.

Happy to move fast on this.