If your H3 Max image-to-video result looks like a photograph, first check whether the video is playing. Then watch the subject against something stationary in the scene. A slow zoom can make a clip feel active even when the action you asked for never happens.
Those two checks cost nothing. They also give you a better next prompt than “more movement, more cinematic.” The question is not whether anything changed. It is whether the right thing changed.
This is a troubleshooting workflow for AI Live Generator's H3 Max Turbo interface. The prompt below is an original suggestion, not a tested result or a promise that one retry will solve the problem.
Is it a still frame, or a playing video of a still subject?
Open the completed result and press Play. Watch the time counter advance, and move through the beginning, middle, and end. Make sure you are reviewing your returned video, not the sample preview shown before a result arrives.
If the player stalls, download the file and try another player before generating again. A result that plays correctly elsewhere does not need another paid attempt. A failed generation with no output is a different problem altogether; there is no motion to diagnose yet.
If the file plays but looks static, pick two landmarks: one on the subject and one in the background. For a person turning their head, compare the direction of the face with the fixed doorframe behind them. For a ribbon fluttering, compare its free end with the knot holding it in place.
Now you can describe the failure precisely:
- Everything gets bigger, but the pose stays the same. The camera appears to move; the subject has not performed the requested action.
- The background changes, but the subject stays put. Rain, smoke, or passing light may have supplied motion without the main event.
- The subject moves, but its shape changes. This is a fidelity problem, not a lack of movement. Asking for even more action may move you further from an acceptable result.
You do not need to identify the model's internal cause to make that distinction. You need to know what the next take must do differently.
Replace the atmosphere with a visible event
“A beautiful, dynamic product shot” describes an impression. It does not say which part of the product moves, how far it moves, or what happens afterward.
For a diagnostic retry, keep the camera still and give one visible part one job. That removes camera movement as a substitute for the action you wanted. It is an instruction, not a camera lock; review whether the result followed it. Our camera-movement guide deals separately with unwanted zooms and drifting framing.
Suppose your starting photograph shows a plain white mug with a blue ribbon tied around its handle. The knot is visible, the loose tail lies on the table to the right, and there is open space above it. You could ask for this:
Keep the uploaded composition and a fixed camera. The mug and the ribbon knot remain in place. One brief breeze lifts the ribbon's loose tail from the table; it bends upward, flutters once, then settles back onto the table before the clip ends. Keep the mug's shape and the tabletop unchanged. Audio: a soft fabric rustle during the flutter, then quiet room tone. No cut, camera move, new object, speech, or music.Use that prompt only with a matching image you own or have permission to use. For another subject, replace the action rather than asking the model to invent this mug-and-ribbon arrangement on top of an unrelated photo.
The knot is the fixed reference. The loose tail is the moving part. Lifting off the table is the event; settling back is the finish. You can judge all three without deciding whether the clip feels “cinematic.”
Choose five seconds in the duration control for this small action. Prompt wording does not set the duration for you. If your real brief needs several movements in sequence, simplify it for the diagnostic attempt; a longer clip creates room, not a guarantee of motion.
Check whether the photograph leaves room for that event
A tightly cropped image may not show the body part you want to move or the place you want it to reach. If the ribbon's end is outside the frame, this is a poor image for judging whether that end lifts. Choose a wider source image before refining the wording again.
Likewise, “preserve everything exactly” conflicts with a request to change a pose. Name the features that should stay stable—the mug's shape and the knot's position here—without telling the model to freeze the whole scene. The image-to-video preparation guide covers source framing and clarity in more detail.
If you supplied a last frame, ask whether it is necessary for this particular test. For an action whose ending composition does not matter, try the first frame alone so there is one fewer endpoint to satisfy. If the ending does matter, keep it and check that both images allow a believable path. Our first-and-last-frame guide explains that separate problem. Do not remove a required endpoint just to obtain more motion.
The upstream Turbo image-to-video schema defines image and end-image inputs. This site's current interface requires a first image and makes the last image optional. It does not offer a motion brush or a motion-strength slider. A prompt cannot turn either missing control into an available feature.
Spend the next attempt on one question
Keep duration, resolution, and Balanced or Quality prompt expansion unchanged while checking a wording revision. Otherwise, an improved result will not tell you which change helped. Even with the same settings, outputs can vary; one better take is useful, but it is not a controlled success-rate measurement.
At publication, five seconds costs 15 credits at 480P or 25 credits at 768P here. Three new five-second 480P attempts use 45 credits. Moving to 768P is another generation, not an upscale of the exact take you already liked. Check the displayed cost and current plans before you submit.
Write one acceptance sentence first: “The ribbon lifts while the mug stays put.” If the next result passes, stop troubleshooting and decide whether it belongs in your project. If it fails, keep the file and name what failed. More adjectives are not a diagnosis—and another generation is only worthwhile when you know what you want to learn from it.
When that sentence is clear, open the generator with your source image and a smaller brief.