- Select which priorities you want to make in your video.
- | Shot type | What to check first | What would make the clip unusable?
- Watch each video in the format they will be seen in.
- If both models are distorting the label during a full rotation of product, reduce the complexity of camera movement to keep the pack in view.
Select which priorities you want to make in your video. Select two of the AI video generators that allow you to include these inputs/outputs and any audio. Use the same prompt and parameters on both models. Compare the videos created by each model. Consider the smoothness, likeness to the input, and cost (credits). If you need consistency in your campaign, you may want to do this for multiple prompts.
When to consider AI video models
| Shot type | What to check first | What would make the clip unusable? |
|---|---|---|
| Text-to-video concept shot | Whether it follows the scene and camera instructions | The main action happens off-screen |
| Image-to-video product shot | Whether it accepts the approved packshot as an input | The label changes or the packaging warps |
| Dialogue shot | Whether the workflow supports the audio you need | The words are wrong or the mouth movement falls out of sync |
| Vertical ad opener | Whether the output fits the intended placement | The product reveal is cropped or comes too late |
Select models to consider from the video model directory. Double-check the exact model variation, inputs, video length, aspect ratio, and audio options in the generator. This will help to write a suitable prompt.

Test 2 models on 1 video
1. Determine what makes a usable video
Decide what should be included in the video, and what needs to be verified by the reviewer. For example, if creating a vertical video for a coffee product launch, have the product in the centre of the frame as the camera slowly zooms in. The roast name should be visible, and the product should stay in shot for long enough to include a CTA.
Example prompt : Create a video from image of coffee bag. Have product upright and facing forward. Zoom in slowly. Ensure label, shape and sealing of bag are similar to original image. Finally, stop the zoom to include end card.
Verification: Check first, middle and final frames match the packshot. If during playback the label is different, pause and note. If the roast name changes at any point, the video is a fail.
2. Use comparable inputs and settings
In both models, start with the AI video generator same input images and prompt. Select the same video length, aspect ratio, resolution and audio options (where possible). Note any differences. If only one model can export in your required format, this may impact your decision.
Generate the same number of takes in each model. If you are creating a campaign with repeated videos, generate 2 takes for each model. If a good take is generated in the first attempt, 1 usable video may be added to the edit. To gain an early understanding of whether you can consistently produce the same result, generate 2 takes.
3. Evaluate generated videos
Watch each video in the format they will be seen in. If generating a Reel, view on a phone to place captions and CTAs. Check off each criterion and compare usable videos.
| Scorecard item | What to record |
|---|---|
| Source accuracy | Pass or fail for the exact details named in the brief |
| Motion and timing | Whether the action happens on cue and the ending is usable |
| Framing and audio | Whether framing survives the final crop and any required audio works in the edit |
| Usable takes | Number of clips that pass all required checks |
| Production effort | Total credits spent and estimated minutes of cleanup |

Consider changing Workflow
If both models are distorting the label during a full rotation of product, reduce the complexity of camera movement to keep the pack in view. If precise wording and packaging are critical, animate the validated image in an editing software with a simple zoom. If the back of the pack needs to be shown, ensure you are using a validated back-of-pack image or video so correct details are visible. Audio considerations
If you have a model that produces the best visuals, but frequently require updates to script, adding a separate voice over means the line can easily be updated without updating the video. Create a new voiceover using your AI audio generator. Check pronunciation, speed and that the line fits well into the edit.

Conclusion
On your next project, test 1 video using 2 compatible models. Decide what would make this video unusable and complete a test. Save the successful prompt, model variation and settings, along with the validated video so your team can reuse this configuration. For inspiration on what to test, look at AI video examples on Linkedin.
Frequently asked questions


