Compare the input, original clip and final version
| Where cropping appears | First check |
|---|---|
| Reference already loses head or hands | The actual submitted image and its shot role; a written plan is not a repaired image. |
| Start is correct but bending or raising an arm leaves frame | Subject movement relative to the camera and available space. |
| Original clip is complete but final output is cropped | The later aspect ratio and crop position. |
Leaving frame is not inherently wrong. Exits, facial close-ups and intentional occlusion can serve the story. Treat cropping as a defect when a required visible fact is lost. A receipt handover needs readable object possession; pulling into an extreme wide shot merely to retain shoes may hide the important hands.
Cover the action while preserving the accepted shot
Video-prompt methods realise accepted starts and ends while inheriting storyboard camera behaviour. A fixed shot specifies framing, height, axis side, movement within the frame and duration. A moving shot specifies trigger, path and endpoint. One interval cannot demand both a fully fixed camera and continuous camera following.
If an accepted facial close-up also requires taking something from below the waist, revise the storyboard or use an approved additional shot rather than silently changing to a wide view inside the prompt. Keep a reference’s composition role separate from end-state facts: an image of a person already holding the ticket cannot serve as the starting state before pickup.
Use the actual revised reference to check the head and important hands. An image not yet acquired is not a frozen keyframe. When planning an input in text, make clear that an actual image is still required for image-to-video submission.
Complete example: keeping the head and ticket hand visible
Original paper shot: at a bookshop counter, Du Qing picks up her own receipt with a blue circular mark and holds it steadily at her chest. It begins on the counter and ownership does not change; no handover occurs. A tight head-and-shoulders plan hides the hand and paper during pickup. The author approves a six-second fixed medium shot preserving the action and final possession.
The creator explicitly chooses text-to-video. This is a composition and motion candidate, without an actual generated clip:
Portrait bookshop-counter scene. Du Qing, an adult woman with short hair and a grey cardigan, stands behind the counter. One white receipt bearing a blue circular mark lies in front of her. Keep the counter, both hands and her head through waist visible, with space above the head. The camera is at eye level on the established left side of the axis and holds this medium framing for six seconds to show the whole pickup and steady hold, without pushing in or panning. She looks down, reaches with her right hand, contacts the paper edge and lifts it. She straightens and holds the receipt at chest height with its blue mark toward the image; her left hand remains on the counter. Head, hand and paper remain visible while bending and straightening. There is one receipt and no transfer. At the end she stands steadily and still holds it in her right hand. Use only quiet shop ambience, without dialogue, voice-over or music. Generate no non-diegetic subtitles, captions or dialogue overlays at any point. The receipt requires no readable text.
There is no universal headroom percentage, and six seconds is this example’s choice rather than a one-try model guarantee. For image-to-video, obtain an actual matching start image with the receipt on the counter, not already at the chest, and bind it according to the target model. Do not reuse a text-to-video preparation statement as proof of an image input.
Check start, middle, end and the final frame format
Watch the complete shot at normal speed, then pause to locate contact, lift, straightening and final possession. Required content should be visible at each point. An occasional clipped head and an entire off-screen action may need different repairs. Check that the following shot still begins with the same receipt in the right hand, without a new hand or duplicate paper introduced by reframing.
If the final edit changes aspect ratio, inspect these moments again. A passing source clip does not prove the cropped version passes. Update only affected shot materials rather than rewriting the series for one framing error. A clearer prompt proposes a repair; actual references and generated results establish whether it worked.
FAQ
Does asking for a full-body shot solve every crop?
No. Choose framing for the shot’s needed head, action and objects, then check actual references, movement range and later cropping.
Do it with the skill
Use $short-drama-video-prompts with the accepted shot, actual input and source clip. Locate reference, movement or edit cropping; retain action and endpoint, updating the storyboard before changing framing. Distinguish text-to-video from real image inputs and inspect the whole action and final aspect ratio.
Read the method: Fixed and moving cameras, framing and relative subject movement · Start/end boundaries, actual references and motion implementation
About Drama SkillsThe short-drama-video-prompts skill on GitHub