Top 7 Trending AI Video Prompts This Week (July 2026)

Every week we collect AI video prompts that creators publish on X and rank them by real engagement. This roundup covers the seven strongest prompts from the past seven days, with the full text so you can copy and adapt them.

The engagement numbers below are the ones recorded when we collected each post. Nothing here is invented, and every prompt links back to the creator who wrote it.

1. A Stone Dragon the Size of a Skyscraper

Seedance — 467 likes, 115,683 views

A stone dragon the size of a skyscraper breaking through clouds. Two figures standing on a cliff watching it rise. Volumetric fog wrapping around carved scales. Light cutting through the clouds behind its head.

This was the most-viewed prompt of the week by a wide margin, and it is also the shortest. Four sentences, no camera jargon, no style stack. What makes it work is scale contrast: a skyscraper-sized creature next to two human figures gives the model an unambiguous sense of proportion, and "light cutting through the clouds behind its head" fixes the lighting direction in a single clause.

If you are used to writing 300-word prompts, this is a useful reminder that specificity beats length.

See the full prompt on the detail page — posted by <a href="https://x.com/cipgerx/status/2079683948452442225" rel="nofollow" target="_blank">Cip⚡️ (@cipgerx)</a>.

2. Forcing Hand-Drawn Animation With Negative Constraints

Seedance — 315 likes, 31,831 views

Style: 8K IMAX, hand-drawn frame-by-frame animation at 12 frames per second, every frame painted by hand, visible brushstrokes, oil-on-paper texture, stop-motion-like temporal stutter, frame-hold animation cadence, hand-drawn line jitter between frames, no smooth interpolation, no 3D render, no game engine, no game-cutscene aesthetic, no CGI smoothness.

The opposite approach to the dragon: a long, heavily structured prompt broken into labelled blocks (Style, Cinematography, Lighting, Atmosphere, Emotional tone, Color, Camera, Physics, Composition, Continuity).

The interesting technique is the run of negatives at the end of the style block. Video models default to smooth interpolated motion, so asking for "hand-drawn" alone usually gets you clean CGI with a painted texture. Explicitly ruling out interpolation, 3D render and game-engine look is what actually produces the frame-hold cadence.

The rest of the prompt is worth studying for the emotional-tone block — a mother and child lost in a blizzard — which drives character performance rather than appearance.

See the full prompt on the detail page — posted by <a href="https://x.com/alex_prompter/status/2078244425734697373" rel="nofollow" target="_blank">Alex Prompter (@alex_prompter)</a>.

3. One Unbroken Shot Through a Forest

Kling — 309 likes, 15,965 views

Hyper-realistic cinematic action sequence using all uploaded images as visual references, presented as one seamless, unbroken continuous shot. A lone motorbike rider begins as a tiny silhouette crossing a glowing meadow beneath an immense, shadowed forest canopy. The camera descends rapidly from a sweeping aerial view, skimming the grass before locking beside the bike as it enters a corridor of colossal moss-covered trees.

This prompt is written as a camera itinerary. Rather than describing a scene, it describes a continuous path: aerial descent, lock beside the bike, swing behind for a low chase, circle the bike, race ahead facing backward, drop beneath a bridge, climb back around.

The closing sentence does the heavy lifting for consistency — it asks the model to maintain the same rider, bike, lighting and forest geography throughout. In a long single-take prompt, that instruction is what stops the subject from drifting halfway through.

See the full prompt on the detail page — posted by <a href="https://x.com/umesh_ai/status/2078377603753554162" rel="nofollow" target="_blank">Umesh (@umesh_ai)</a>.

4. A Transformation Sequence With a Fixed Ending

Seedance — 370 likes, 24,402 views

Cinematic 15-second video of a young woman with long blonde wavy hair transforming into a modern street-style Spider-Woman in New York City at golden hour.

Transformation prompts are consistently popular, and this one is structured well: the wardrobe is fully specified up front (black bomber jacket, pink tights, grey sweater tied at the waist, white gloves, teal sneakers), then the action beats run in order — sitting on a ledge, shooting a web, swinging through rainy streets, landing on a rooftop.

The last beat is stated as a "final shot": standing on a water tower against an orange sunset. Naming the closing frame explicitly is a reliable way to stop a video model from ending mid-motion.

See the full prompt on the detail page — posted by <a href="https://x.com/Just_sharon7/status/2080641010393354560" rel="nofollow" target="_blank">Sharon Riley (@Just_sharon7)</a>.

5. A Chase Scene Written as a Shot List

Seedance — 259 likes, 30,986 views

A cinematic 10-second realistic video of a terrified young woman [...] desperately running barefoot through a dense, misty prehistoric jungle. A massive, photorealistic Tyrannosaurus rex with detailed scaly gray skin and open roaring jaws aggressively chases her from behind and to the side.

What separates this from a generic monster prompt is the explicit camera plan in the middle: tracking shot following the runner, then over-the-shoulder of the T-Rex, then a close-up of muddy bare feet splashing, ending on an extreme close-up of her face.

That is four shots inside a ten-second clip. Listing them in order gives the model an editing rhythm instead of one long static chase, and the closing reference to a well-known film look sets the grade without needing colour terminology.

See the full prompt on the detail page — posted by <a href="https://x.com/Just_sharon7/status/2080539029985198518" rel="nofollow" target="_blank">Sharon Riley (@Just_sharon7)</a>.

6. Mechanical Assembly, Described Part by Part

Seedance — 234 likes, 9,989 views

High-tech mechanical suit components rapidly assemble onto her body, starting from her chest and spreading outward. A purple and black powered exoskeleton helmet locks into place over her head, her eyes glowing red upon activation.

The useful detail here is directional: the assembly starts at the chest and spreads outward, and the helmet locks last. Mechanical-assembly prompts fail when the model has no ordering to follow and simply cross-fades between the before and after states.

The prompt then pulls the camera out for a landing pose, which gives the sequence somewhere to resolve — the same closing-frame trick as the transformation prompt above.

See the full prompt on the detail page — posted by <a href="https://x.com/AvelyrahnAI/status/2077754340543283508" rel="nofollow" target="_blank">Avelyrah (@AvelyrahnAI)</a>.

7. Putting Brand Text Into the Frame

Hailuo / MiniMax — 436 likes, 26,716 views

Cinematic classroom portrait of a beautiful young woman in a school-style uniform sitting on a desk, barefoot. She wears a white short-sleeved shirt with 'Keor' elegantly embroidered on the chest (replacing original text) and a red-toned plaid skirt.

This was the most-liked Hailuo prompt of the week, and the reason to include it is the parenthetical: "(replacing original text)". Embedding a specific brand word into clothing is one of the harder things to control, and the instruction to replace rather than add stops the model from rendering both the invented logo and whatever generic text it wanted to place there.

The same creator posted several variations this week using the same replacement technique on outfits, embroidery borders and skin, which is a reasonable signal that the pattern is repeatable rather than a one-off lucky generation. If you are prompting for Hailuo specifically, our Hailuo H3 prompt guide goes deeper on how the model handles motion and camera direction.

See the full prompt on the detail page — posted by <a href="https://x.com/KeorUnreal/status/2079225480021184687" rel="nofollow" target="_blank">KeorUnreal (@KeorUnreal)</a>.

What This Week's Prompts Have in Common

Across all seven, three patterns repeat:

• The ending is stated. The strongest prompts name a final shot or pose. Video models are much more likely to resolve cleanly when they know where the clip stops.
• Order matters more than adjectives. The chase, the transformation and the suit assembly all sequence their beats explicitly. Descriptive detail without ordering tends to produce a static scene with motion sprinkled on top.
• Negatives are used deliberately. The hand-drawn animation prompt rules out the model's defaults instead of hoping a positive description overrides them.

Notably, prompt length was not a predictor of engagement. The most-viewed prompt of the week was four sentences long; the second was several hundred words. Both were precise about the one thing that mattered most to the shot.

Browse More Prompts

We collect these prompts continuously and organise them by model. You can browse the full library on trending prompts, filtered by Seedance, Kling, Hailuo, Midjourney, Veo, Runway and more.

If you want to work the other direction — starting from a video you like and recovering the prompt behind it — try our video to prompt tool. Paste a YouTube link or upload a clip, and it returns a structured prompt describing the scene, motion and camera work, which you can then adapt for the model you are using.

New roundup every week.