Awe in a spectacle shot comes from a comparison, not from the word "giant". A colossal figure is only colossal next to something whose size the viewer already knows, and the camera decides which of those things ends up nearest the lens. This article builds a scale-contrast spectacle shot in CapCut at 16:9, a stone titan in the sea beside a packed stadium, four stills and two eight-second clips, and measures each still the way a viewer's eye does: the giant's height on screen against the nearest human-sized thing in the frame. The first result is the reason the article exists.
The prompt that asked for the most got the smallest giant
The first still was written the careful way, with the camera placed, the giant's proportions tied to the stadium and the people tied to its toes.
Photoreal cinematic still, 16:9 landscape, night. An open-air seaside stadium packed with a crowd holding up phone lights. The camera is down among the crowd at head height, looking up past the backs of the people in front. Beyond the far stand a colossal figure stands in the sea: a titan of weathered dark stone and green bronze, seaweed hanging from its shoulders, a plain faceless helm, holding a long trident upright. Its ankles are level with the top tier of the far stand and its head is high above the middle of the frame; the people in the front rows reach only to the height of its toes. Stadium floodlights rake across its legs, sea spray drifts down through the light. No text, no logos, no real landmarks, no flags.
It came back as a handsome picture and a small giant. The titan stands in the flooded pitch rather than in the sea beyond the stand, and on screen it is 470 pixels tall in a 720-pixel-high view of the frame while the head and shoulders of the spectator nearest the camera are 200 pixels: the giant is a little over two heads high. Nothing in the prompt was ignored so much as outweighed. "Down among the crowd at head height, looking up past the backs of the people in front" put a human-sized ruler right against the lens, and a ruler that close makes anything behind it small. The sentence about ankles and top tiers was true in the picture and beside the point on the screen.
Measure the giant against the nearest thing you know
The measurement is simple enough to do on any result: the giant's height on screen divided by the height on screen of the nearest thing whose real size the viewer knows, usually a person. It is not the giant's "real" size; it is the number the picture actually shows. Three more stills were made to move that number, each at one credit through the CapCut AI image generator flow in Video Studio's Image mode at Seedream 4.5, 16:9 and 2K, and all four are measured below.
The same titan measured four ways. Red is the giant on screen, yellow the nearest person on screen; the ratio is what a viewer's eye is given to work with.
The numbers run 2.3, 3.9, 11 and 15, and the order is the argument. A camera in the crowd gives a ratio of 2. A prompt with no camera position and no references at all gives 4, because the model chose its own framing and put only the upper half of a different, bearded titan in the sea, with the stadium cut off at the right edge. An aerial view gives 11, with the whole stadium as the ruler and the people reduced to dots. A camera on the ground with nothing in the foreground and a row of people standing at the titan's feet gives 15, the highest of the four, with the whole body in the frame from helm to toes. The wording about ankles and tiers was identical in the first, third and fourth prompts; only the camera and the foreground changed.
Put the ruler at its feet and the camera on the ground
The fourth still is the shot to keep, and the two phrases that made it are short. "The camera sits on the ground at the edge of the empty pitch, very low, looking up, with nothing and no one in the foreground" removes the near ruler. "The people in the stands are tiny next to its feet" puts a far ruler where it can be read against the giant at the same depth. The model added the row of spectators standing on the shoreline at the titan's toes on its own in this run; it is the best thing in the picture, because the eye reads those figures and the foot beside them in one glance.
Photoreal cinematic still, 16:9 landscape, night. An open-air seaside stadium packed with a crowd holding up phone lights in the stands. The camera sits on the ground at the edge of the empty pitch, very low, looking up, with nothing and no one in the foreground. Beyond the far stand a colossal figure stands in the sea: a titan of weathered dark stone and green bronze, seaweed hanging from its shoulders, a plain faceless helm, holding a long trident upright. Its ankles are level with the top tier of the far stand and its helm reaches the top of the frame; the people in the stands are tiny next to its feet. Stadium floodlights rake across its legs, sea spray drifts down through the light. No text, no logos, no real landmarks, no flags.
Two things about the frame follow from the ratio. The feet have to be in it, because the ruler has to stand next to something, and a giant cropped at the knees loses the one comparison the shot is built on. And the top of the helm should touch or leave the frame, because a giant with sky above its head is being told by the frame that it fits. The fourth still does both; the first still did neither, with sky above the helm and the feet standing in water at the level of the crowd's phones.
What the word giant does on its own
The plain prompt, "a giant stone titan with a trident stands in the sea beyond a seaside stadium, and the crowd in the stadium watches it", is worth keeping as a control because it shows what the model supplies when nothing is decided for it. In this run it supplied a high camera on the stands, a bearded titan of cracked stone rather than the helmeted one, a waist-deep pose in the surf, and a framing that cuts the stadium at the right edge and the titan's legs below the waterline. The ratio is 3.9, respectable, but the picture has no feet, no top and no full stadium, so it cannot be measured by a viewer either; it looks big in the way a close-up of anything looks big. The word did not fail. It left the two decisions that make scale readable, where the ruler is and where the camera is, to chance.
Big things move slowly in a spectacle shot
Scale has a second reading in video, and it is speed: a body that big raises an arm over seconds, not frames. The first still, the one with the crowd camera, was attached to the composer for the CapCut AI video generator in Video mode with the settings panel set to Video clip, Seedance 2.0 Mini, 720p, 16:9 and 8s, and this prompt.
The camera does not move. Over the first four seconds the titan slowly raises the trident above its helm, its movement heavy and slow, the head barely turning. Behind it a wall of seawater rises higher than its shoulders and hangs there, spray drifting down through the floodlights. The crowd's phone lights sway and flicker. Nothing is destroyed and no one is hurt; the stadium stays as it is. No cut, no zoom, no pan.
The Video mode settings for the clip: 16:9 and 8s, Seedance 2.0 Mini at 720p.
The cost lines: 1 credit for each still at Image mode, Seedream 4.5, 16:9 and 2K; 64 credits for the eight-second clip at the settings above, on 8 September 2026.
On 8 September 2026 the chat printed "Generating 1 item will consume 1 credit" before each still and "Generating 1 item will consume 64 credits" before the eight-second clip at Video clip with Seedance 2.0 Mini, 16:9, 8s and 720p. Those lines are the only credit figures used here; the five-second and ten-second durations, other models, resolutions and modes are priced differently, so read the line before you send.
The eight-second clip from the crowd-camera still, as downloaded: 1280 by 720, 24 frames per second, 8.04 seconds. Turn the sound on for the clip's own track. Shown here as frames in sequence; the full video and a GIF version are supplied alongside this document.
The clip in nine frames. The trident takes about five seconds to go from resting to raised; the water is up by the fourth second and stays.
The raise took about five seconds, from the second second to the seventh, which is slow by the standards of a five-second clip and right for a body this size. Measured as change between consecutive frames inside the titan's part of the picture, the motion runs at about 1.6 in the first second, peaks near 2.4 in the third and fourth, and falls to 1.0 in the last; the foreground crowd, phones and all, stays between 0.7 and 1.2 throughout. The wall of water rose higher than the prompt asked, above the far stand and over the titan's shoulders, and hung there; in this run it became a second ruler, and a better one than the crowd, because it stands at the titan's depth. The clip carries an audio track of its own, a steady bed at about minus 17 dBFS with a third of its energy below 150 Hz, a low rumble with no strike or crash in it; nothing was added from the editor's library. The stadium is intact in the last frame, as the prompt required. What the clip inherits from the still is the ratio: this is a moving giant that is still a little over two heads tall on screen, because the camera did not leave the crowd.
The fourth still was then given the same prompt at the same settings, and it took three requests to get a clip back: the first two posted the message with the attached still and then sat on "Generate videos" with no progress and no video card for more than ten minutes each, with no cost line on one and a cost line but nothing after it on the other; the third ran normally. If no progress appears after several minutes, check the task status and connection. Retry only after confirming that the original request did not begin processing or consume credits.
The same prompt from the ground-camera still: 1280 by 720, 24 frames per second, 8.04 seconds. This is the clip to keep. A GIF version is available.
The ground-camera clip in nine frames. The people at the titan's feet stay in every frame, so the ratio the still had is the ratio the clip keeps.
Here the raise is done in four seconds, as the prompt asked, and the wave arrives on time: the sea at the left of the frame changes hardly at all in the first two seconds, then jumps as the wall builds from the third, and the titan's part of the picture moves at about 2.7 per second through the middle of the clip against 1.9 at the start and 1.6 at the end. The soundtrack it came with rises from about minus 26 dBFS to minus 15 as the water builds and eases off at the end, a swell rather than a crash, with a quarter of its energy below 150 Hz. What makes it the better clip is not the motion, which is close in both; it is that the ruler survived. The row of people at the titan's feet is in every one of the nine frames, so the fifteen-to-one reading of the still is what the viewer keeps watching, while the crowd-camera clip spends eight seconds confirming that its giant is two heads tall.
Keep the feet and the ruler in the frame
The rule the four stills and the clip agree on is a framing rule rather than a wording one. Write the giant however you like, but put the camera where nothing human-sized is close to it, put people or a building at the giant's feet at the same depth, keep the feet in the frame and let the head reach the top. Then measure the result the way this article did, giant pixels over nearest-person pixels, and if the number is under 10, move the camera before you touch the adjectives.
What was not made here, by choice: no real landmark, no real city, no destruction of the stadium or its crowd, and no figure drawn from any religion's iconography; the titan is a helmeted stone-and-bronze design with a trident because a trident reads as scale against a body. Sound effects from the editor's library were not added; the clip's own audio track is described above and left as it came. The next shot to make is the same measurement applied to your own giant: one still on the ground with a ruler at the feet, read the ratio, then spend the video credits.
Written 8 September 2026. All stills and the clip are generated images and video from CapCut made for this article; no real place, person, brand or religious figure appears. Interface labels, panel contents, the cost lines and clip specifications reflect a single session on that date and the configurations named above, and may change. Screen ratios were measured on the downloaded files scaled to a 1280 by 720 view; the boxes in the figure show what was measured.