Blooginga
A few months ago, a freelance client asked me to turn a single product photo of a perfume bottle into a 10-second ad. The budget was tight, the deadline was Friday, and the only thing the agency had given me was one flat-lit JPEG.
In the old world, that meant three days in After Effects, a stock music license, and a prayer. In 2026, the same deliverable takes about an hour — if you pick the right AI video pipeline. Here is the workflow I now reach for, and the mistakes I made before I got it dialed in.
The tooling landscape right now
There are roughly four categories of AI video tools, and you usually need two of them to finish a real ad:
Image-to-video engines: animate a still image with camera moves and subject motion. Best for product shots, portraits, static scenes.
Text-to-video engines: generate motion from a written brief. Best for establishing shots, cinematic B-roll, abstract concepts.
Motion-control models: you provide a reference video (a camera path, a gesture) and the engine applies that motion to a new image.
Image-to-image editors: used upstream to re-style, re-light, or re-cut a frame before you feed it to the video engine.
Most platforms force you to pick one. A small number wrap all four into a single workspace. Movoria Studio (Movoria Studio) is the one I settled on, mainly because I can move from a text prompt to a finished clip without copying assets between five browser tabs.
A worked example: perfume bottle in motion
Starting image: a flat-lit JPEG of a black perfume bottle on a white seamless.
Step 1 — re-style the source frame. Image-to-image with a "cinematic, low-key, golden rim light" prompt. The bottle now looks like a hero shot instead of a catalog photo.
Step 2 — animate with image-to-video. Prompt: "Slow dolly-in, bottle stays still, smoke drifts across the frame, soft golden rim light." The image2video model keeps the product geometry locked while adding motion to the background and particles.
Step 3 — finish with a text2video opener. For the first two seconds, I generated an abstract "scent" visual from a text brief: "Slow-motion macro of amber liquid dispersing in water, warm backlight, 24fps, cinematic grain." Cut it in front of the product shot.
Total time: 45 minutes. Two models, one workspace, one credit system.
What went wrong the first time
Before I learned to do steps in the right order, I tried to go text-to-video directly for the whole ad. The model invented a bottle that looked nothing like the client product. Lesson: always drive video from a real image when the product has to be on-model. Save text-to-video for mood, openings, and B-roll.
Cost and credits
Without naming exact prices (they change), the broad shape is: text-to-video costs more per second than image-to-video, and motion-control costs the most. A 10-second ad end-to-end on a mid-tier plan usually runs in the single-digit credits, which is a fraction of what stock footage plus editing time costs.
When not to use this
If your client needs broadcast-grade 4K with strict color science, AI video is not there yet. You can use it for the moody 70% and finish the last 30% in a traditional compositor. If you are making a 30-second spot, you will still want human editing. If you are making 50 social shorts per month, this is the workflow that scales.
Try it
Movoria Studio has a public playground at Movoria Studio where you can test the image2video, text2video, and motion control models on your own images before you commit. I would start with a single product photo and three different motion prompts — that test alone will tell you whether the rest of the pipeline is worth building on top of it.
This article is not sponsored; the author is an independent user of the tool. [Visit Movoria Studio](Movoria Studio)
Learn more: Movoria Studio