I Migrated 200 Product Shots from Sora to Veo 3.1: Here's What Broke
AI-generated product photography is the practice of creating commercial product imagery through text-to-image and text-to-video models rather than traditional camera setups. This matters for ecommerce sellers because image quality directly affects click-through rate, conversion rate, and ad spend efficiency on platforms like Amazon, Shopify, and TikTok Shop.
When OpenAI's Sora dropped its still-image pipeline and Google countered with Veo 3.1, I had 200 product shots sitting in a Sora render queue. Migrating them cost me a week, a small fortune in API credits, and several lessons I am going to save you from learning the expensive way.
What Broke When I Pushed 200 Shots Through Veo 3.1
Use this section as directional guidance. Validate the claim against your own catalog data, product samples, and channel requirements before publishing or scaling the workflow.
The second problem was color science. Sora leaned warm, Veo 3.1 leans clinical. Three of my skincare products, all in the same neutral beige palette, came out looking like they belonged in a hospital supply catalog. Sora's color grading was closer to real studio lighting with a softbox bounce. Veo 3.1 needs explicit prompts to mimic that.
The 4 Specific Failures I Logged
Every failure is a prompt-engineering lesson. Here is what the broken renders taught me.
1. Material misreads on reflective surfaces
Chrome, polished steel, and mirrored acrylic are Veo 3.1's weakness. Sora handled brushed metal with a single diffuse reflection. Veo 3.1 hallucinates additional light sources, which is great for video atmosphere and terrible for catalog imagery. I had to add a negative prompt: "no extra light sources, no environmental reflections, softbox only."
2. Text and label degradation
Use this section as directional guidance. Validate the claim against your own catalog data, product samples, and channel requirements before publishing or scaling the workflow.
3. Skin tone inconsistency in fashion
A batch of 40 apparel shots came out with inconsistent skin tones between the model and any visible hand or arm. Sora was slightly better here because it had more training on natural light photography. Veo 3.1 needs explicit skin tone references in the prompt or it averages to a "studio neutral" that does not match real demographics.
4. Aspect ratio cropping for marketplace specs
Amazon requires 1000x1000 minimum on a 1:1 ratio. TikTok Shop wants 1080x1920 vertical. Sora handled both in one pass. Veo 3.1 needed a separate render per aspect ratio, which doubled my generation time and my API costs.
What I Wish I Had Known Before Day One
The biggest lesson: generative video models are not drop-in replacements for image models. They are designed for motion, narrative, and temporal coherence. Product photography is the opposite problem. You want stillness, neutrality, and reproducibility. That gap is where most of my broken renders came from.
A video model makes a beautiful 5-second clip of your product rotating on a turntable. It does not make a 1000x1000 catalog hero image. Different tool, different job.
For the actual job of product imagery, a clean white background, consistent lighting, accurate color, and predictable framing, purpose-built tools outperform general video models. Sora was already a compromise. Veo 3.1 is a bigger compromise, just with newer technology behind it.
The Workflow That Actually Worked
After burning through the 200 Sora shots and the failed Veo 3.1 re-renders, I rebuilt the pipeline around a specialized AI product photography studio for ecommerce listings for the base imagery, then reserved Veo 3.1 for short-form video ads only.
Here is the workflow I now run for any new product drop.
Step 2. Use a dedicated AI background remover for product images to clean up any edge artifacts before export.
Step 3. Reserve Veo 3.1 for short ad creatives (6-10 second vertical videos) where its cinematic motion is actually a strength.
Step 4. For mockups on packaging, apparel, or print, layer a dedicated mockup generator for ecommerce product pages instead of fighting a general video model.
This split keeps each tool in its lane. The still images come out predictable and marketplace-ready. The video ads get the cinematic quality that Veo 3.1 was actually trained to produce.
Veo 3.1 vs. Sora vs. Purpose-Built Tools
Migration Checklist for Your Own Pipeline
- Audit your existing Sora outputs for color, framing, and material accuracy
- Re-render only the failed shots, not the full 200, to control API spend
- Test Veo 3.1 on 5 products before committing to a full migration
- Reserve Veo 3.1 for video ad creative, not still catalog imagery
- Use a purpose-built product photo tool for hero shots and white-background cutouts
- Run a final color check on a calibrated monitor before listing upload
Frequently Asked Questions
Is Veo 3.1 actually worse than Sora for product photography?
Not worse in raw capability, just worse fit. Veo 3.1 is a video model that can produce stills as a byproduct. Sora's still-image mode is closer to a dedicated image model. For catalog work where stillness, color accuracy, and aspect-ratio precision matter, both are compromises. A purpose-built AI product photography tool outperforms both for the specific job of marketplace imagery.
How much did the 200-shot migration cost in API credits?
Use this section as directional guidance. Validate the claim against your own catalog data, product samples, and channel requirements before publishing or scaling the workflow.
Should I use Veo 3.1 for ecommerce product videos?
Yes, that is where it earns its keep. Short vertical ad creatives, rotating product turntables, and lifestyle b-roll are exactly what Veo 3.1 was trained for. Use it for motion work and use a dedicated still-image tool for your catalog hero shots, white-background cutouts, and marketplace listings.
Skip the Migration Headache
Generate marketplace-ready product shots in 1:1, 4:5, and 9:16 with consistent color and lighting. No video model workarounds required.
Try Rewarx Free