Anyone here solved AI creatives for awkward/dynamic eCom products?

Hey everyone, this is my first post here. Happy to have found the forum.

I’ve been working on a bunch of AI automations for eCom image/video generation using both ComfyUI with open-source models and API calls with closed models.

I’m getting decent consistency on simple static products — stuff like pill bottles, creams, etc. Character consistency is also coming along pretty well.

The part I’m still trying to solve is dynamic products or weird-shaped products. For example, electronic neck massagers, posture correctors, and similar items that don’t behave as nicely as standard product shots.

Has anyone here had real success with those kinds of products? Especially for video creatives or multi-angle consistency?

Happy to share what I’m testing on my side too — LoRAs, models, workflows, prompts. Keen to see how far this creative freedom can go.

Which models are you using? Nano Banana Pro (Gemini 3 Pro Image Preview) should handle it pretty well. For videos veo-3.1.

Video side: at the moment I’ve been testing LTX 2.3 in ComfyUI to see how far I can push it with LoRAs and that side of things, and I’ve also been using Kling 3.0 through API workflows. I actually haven’t tried Veo 3.1 yet, so I definitely need to give that a shot.

On the image side, I’ve been using Z-Image Base and Turbo in ComfyUI as well. Still need to get some API automations set up there properly though haha.

How good is Veo 3.1 with dynamic products though? Like can you upload solid reference images and get outputs that are accurate enough for real product demo use?

I’ve also been thinking about using Wan 2.2 more as a character/background replacement layer, where the actual product demo stays unchanged haha )

You’re running into a real limitation — these products are hard because models don’t understand their structure well.

What works:

• Use ControlNet + IP-Adapter in ComfyUI → control shape, don’t rely on prompts
• Train LoRAs with diverse angles, not just clean shots
• For video: keyframes → interpolate (try AnimateDiff)
• Break products into parts + inpainting
• Start from real reference images