First and Last Frame Control
Direct a video model between two fixed images so the motion, timing, and transition land exactly where you need them for an edit.
[Attach your first frame image, and the last frame image if your tool supports it, then use this prompt:] Animate from the first image to the last image. What changes between them: [THE ONE TRANSFORMATION, e.g. "the closed box opens and the product rises out of it" / "the empty street fills with morning light" / "the character turns from facing away to facing camera"] **Motion**: [WHAT MOVES AND HOW, in physical terms: "the lid hinges back over about one second, then the product rises smoothly on a vertical path"]. Everything not named here holds still. **Camera**: [STATIC / SLOW PUSH IN / SLOW PULL BACK / GENTLE ARC LEFT]. One camera move only, at a constant speed, with no cuts. **Timing across [DURATION] seconds**: the change begins at about [X] seconds, completes by about [Y] seconds, and the final second holds on the last frame with only [SUBTLE RESIDUAL MOTION, e.g. a slight settle, drifting dust, a flag still moving]. **Hold constant**: the lighting direction and color temperature, the palette, the background, the framing, and the identity and wardrobe of anything in shot. Nothing appears that is not in either image. **Physics**: motion is weighted and continuous, with no teleporting, no morphing between shapes, and no elements that vanish and reappear. Objects obey gravity and contact. Audio: [SFX AND AMBIENCE, or "no audio"]. Clean frame with no text overlays, no captions, and no watermark. --- If your tool takes only a first frame, replace the last-frame reference with an explicit description of the end state: "By the final frame: [DESCRIBE EXACTLY WHAT THE LAST FRAME SHOULD LOOK LIKE, in the same terms as the first: subject position, camera distance, light, and what has changed.]"
How to use
Fixing both endpoints is the most reliable control available in AI video, because the model only has to solve the motion between two known states rather than invent a destination. The hold-constant list is what keeps generated clips cuttable together, since the usual failure is not bad motion but a wardrobe or light shift midway. Keep the final second static: a clip that is still moving at the last frame is very hard to cut out of.
More video prompts
Generate a sequence of shots featuring the same character. **Step 1: write the character lock.** This exact text goes into every shot prompt, word for word, before anything else. Never paraphrase it. CHARACTER LOCK: "[NAME], a [AGE] year old [DESCRIPTION: build, height impression], with [HAIR: exact length, color, texture, and how it is
Multi-Shot Character Consistency
Keep one character looking like the same person across separately generated shots using a locked description block and fixed style spine.
Write and shot-list an explainer video. What we are explaining: [PRODUCT, FEATURE, OR CONCEPT] Who is watching and what they already know: [AUDIENCE + BASELINE] The one thing they should understand afterward: [THE SINGLE TAKEAWAY] What they should do next: [THE ACTION] Length: [60 / 90 / 120 seconds] Visual style: [SCREEN RECORDING / MOT
Explainer Video
Turn a product or concept into a 60 to 90 second explainer with a script timed to shots, so the visuals carry what the words cannot.
A repeatable pipeline for a faceless channel. Run each step, check the output, then paste it into the next. **Step 1: The script.** ``` Write a [60 / 90 / 180] second script about [TOPIC] for a [NICHE] channel. Audience: [WHO WATCHES AND WHY] Channel voice: [e.g. "calm and factual" / "fast and irreverent"] Format: [top-list / single st
Faceless Channel Pipeline
A four-step chain from topic to finished faceless video: script, shot list, generation prompts, and the metadata to publish it.