Text to video generation for a shark surprise in the middle of a water slide. Run locally using ComfyUI and MiniMax H3, stitched together with Shotcut.
I played around with a few options to try to make this clip the best possible but I still settled for quite a few goofs. Since it was just for fun, it wasn’t worth further attempts to fix it. I tried one render of the whole thing, which worked but then made it more painful when something just needed a small tweak. The final clip ended up being three 10-second clips stitched together (and even a fourth clip for a better segue between the first and second clip). I used the first frame and last frame parameters to help keep the AI consistent across the clips.
I initially wanted to have the rider do an introduction like he was a YouTuber that reviews water slides. A few renders had clips that started at the top but eventually they were already starting in motion. I could have tried a separate clip for the introduction but decided against making it more complex.
I liked the animation of the water and the motion in the slide. I liked some of the laughter and excitement in the voice. I dislike some of the gibberish where the person is talking when “No spoken language” was specified in the prompt. The sharks and the paddling could have been better. The shark bite in an inner tube might be a flaw in the concept. I really meant for it to be deflated, but I thought it was still funny and the ability to see through it at the end was kind of neat.
Final clip:
Prompt#
Style: Hyper-realistic YouTube vlog, GoPro chest-mounted POV, wide-angle lens, shaky handheld motion, 4k, high-fidelity water physics, natural lighting. Subject: A man in his 30s wearing a dark blue swim shirt and blue shorts, barefoot, riding a bright yellow innertube. As the camera is on his chest, only his arms and legs are seen when they pass in front of the camera. No spoken language. Textures: concrete and plastic
Action Sequence: The video begins with a first-person POV of a rapid descent down a closed, twisting waterpark slide with water splashing against the camera. The scene adjusts from all black as eyes adjust to being in the tunnel. Movement slows significantly. Man is laughing. Three shark dorsal fins appear randomly and circle around the innertube. Man starts to panic; he paddles the innertube frantically toward the dark tunnel in front of him. Camera fades to black as he enters the next sequence of enclosed waterpark slides and continues in an enclosed waterslide tube with water splashing against the camera. Rapid twists and turns with innertube going high up against the walls with the curves. At the end, the tube opens up and man splashes into sunlit pool. He stands up, hold the backside of the tube up to the camera where a shark bite in the inner tube is seen.
Storyboard: 00-08s The Initial Descent Man on yellow innertube drops into a closed slide. Rapid twists, water splashing against the lens, concrete/plastic textures. High-speed POV, shaky GoPro motion. 08-12s The Pool Exit Slide opens into a bright, sunlit 50ft pool. Movement slows significantly. Man is laughing. Transition from dark/enclosed to high-brightness. 12-18s The Shark Encounter Two shark fins appear, circling the innertube. Man’s expression shifts to panic; he paddles frantically toward a dark tunnel. Fast panning/swaying as he paddles. 18-25s The Dark Tunnel POV enters a dark, twisting pipe. Periodic sunbeams pierce through seams. Rapid motion resumes. Low-light, flickering sunbeams, high-speed. 25-30s The Finale Exit into a final sunlit pool. Man stands up, holds the tube up. Close-up on a jagged shark bitemark on the yellow rubber. Sudden brightness, macro-focus on the bitemark.
