Crazy Frog AI Prompt Guide: Keep the Motorcycle Invisible and the Rider Consistent

In our countryside sample, the rider travels with bent knees while the pursuer has a white hover bike. The space beneath the rider stays empty. The flip briefly breaks that seated silhouette, then the rider returns to the pose and pulls away. Look at the whole body and the gap behind it, not just the face.

Those are the visual constraints the example prompt below tries to express. The embedded clip is an actual output of our prepared countryside template. It was not generated by submitting the public example prompt in this article, so it should not be treated as a test of that text.

Use this original example only in a tool mode that supports the references you supply. It assumes a chase video you may use, one rider image and, optionally, one chaser image. Replace the bracketed names with the labels recognized by your tool. If you omit the chaser photo, retain the reference clip’s pursuer rather than mentioning a missing Image 2.

This is an editable starting point, not our production recipe or a tested universal fix. Keep the source clip’s supported duration and framing. Do not ask for a flip if your chosen reference never shows one, or assume a model accepts video and two images simply because another mode does.

Check whether your instructions mostly describe a motorcycle or mostly describe a person’s posture. The rider should have a lowered hip position, bent legs, feet clear of the ground and empty hands extended forward. Naming the object and adding the word invisible leaves much more for the model to interpret.

Make the prohibition local to the rider. Banning all wheels or all chairs from the entire scene conflicts with a chaser who is meant to ride one. In the countryside, the pursuer’s hover bike belongs behind the rider; in the office, a rolling chair belongs under the chaser. Neither is a support for the foreground person.

If the person runs instead, review the feet and hips in the reference. A reference dominated by a running gait asks for a different movement. Simplifying the text cannot be assumed to override incompatible source motion. Change one variable at a time so you can distinguish a pose problem from a casting problem.

Use one person per photo and bind the tool’s actual reference labels to rider and chaser. Do not rely on upload order alone. Describe the rider as the foreground person moving ahead and the chaser as the separate person behind. Avoid requesting extra bystanders or an additional passenger while you are checking those identities.

Inspect the face before the tuck, immediately after the flip and as the rider approaches the camera. The turn and temporary occlusion are where you may notice a mismatch. A sharp portrait supplies better reference information, but prompt wording does not guarantee the face will remain unchanged through fast motion.

In our office result, the blue-shirt rider moves through a glass corridor while the second character follows on a rolling chair. The longer eight-second clip gives the escape room to continue down the hallway. The visible shirt is an appearance cue from the rider portrait; the supplied scene defines the corridor and chair.

If you adapt the prompt, align the chaser’s prop, ground surface and camera path with your reference rather than replacing only the word countryside with office. This embedded sample comes from VidMello’s own office setup, not from running the text above with a location word changed.

If you want either of these two prepared scenes, use the template instead: one adult rider portrait, an optional separate chaser portrait and the selected scene. It uses 10 Credits for a six-second countryside or eight-second office video in landscape 16:9 at 720p. You do not upload a motion clip or paste this prompt into VidMello.

The demo music is added after generation. A sentence about matching the beat does not give our preset audio-driven choreography, and the flip’s timing may vary. Review the whole result before deciding on another paid attempt; a clearer prompt or photo is not a guarantee of improvement.