Head to head: Happy Horse Video Edit vs Heygen v5 Digital Twin
Happy Horse Video Edit vs Heygen v5 Digital Twin
By Ryan Merket · Published · Updated
This matchup tests scene adherence and temporal consistency across pastry-making, a rainy street, a tandoor kitchen, and a living-room lighting transition. The assessment is limited to this saved task sample.
The gap was not a matter of polish. Across the pastry, rainy-street, tandoor, and lighting-transition prompts, Happy Horse Video Edit generally depicted the requested subject, setting, and action. Heygen v5 Digital Twin repeatedly returned a seated interview scene instead, including on prompts that called for cooking, rain, or changing light. The saved aggregate scores are **32.2** for Happy Horse Video Edit and **4.7** for Heygen v5 Digital Twin. Happy Horse won all **12** task outcomes, with **0** wins for Heygen and **0** ties, across three successful repeated trials. Happy Horse did have some prompt details that were not clearly shown, such as the full pastry-finishing sequence, but those omissions did not erase the much larger difference in task fit. **Verdict: Happy Horse Video Edit is the clear result on this task sample—not a statistical claim.** This assessment applies only to the saved tasks and does not establish general capability superiority.
Blue Apron Mille-Feuille
A 7-second single unbroken shot in a quiet pastry studio: Mara Vell, a freckled pastry chef with a blunt black bob, silver hoop earring in her left ear, and a cobalt-blue apron over a cream pinstriped shirt, stands at a marble bench steadily assembling a raspberry mille-feuille—piping vanilla cream, placing three precise pastry layers, and finishing with powdered sugar—while the camera performs a slow fluid gliding gimbal arc from front-left to front-right at chest height; soft north-window daylight and a warm practical lamp create gentle highlights on the pastry and her face, and her appearance, clothing, hairstyle, and identity must remain perfectly unchanged and stable from first frame to last with no morphing, flicker, or accessory changes, mood calm, meticulous, and intimate, 16:9
OpenAI: GPT-6 Luna: Model A closely matches the pastry-studio scene and chef, showing cream piping, raspberry pastry, and a consistent appearance across changing camera angles. Model B is a stable, polished interview-style shot but does not depict the requested chef, setting, or pastry action. (Second judge pass, order swapped — scores are the average of both: Model B shows a seated interview scene unrelated to the pastry-making prompt. Model A depicts the specified chef piping cream at a pastry bench, with the req | Anthropic: Claude Fable 5.1: Model A closely follows the prompt with a freck…
Temporal consistency
A man in a yellow raincoat walking toward camera down a rainy street; his face, coat, and umbrella must stay perfectly consistent with no morphing or flicker from the first frame to the last, steady tracking shot, 16:9.
OpenAI: GPT-6 Luna: Model A closely follows the prompt with a man in a consistent yellow raincoat and umbrella on a rainy street, maintaining a stable composition across the frames. Model B shows a woman seated indoors, substantially departing from the requested subject and setting. (Second judge pass, order swapped — scores are the average of both: Model B shows a woman seated indoors, so it misses the requested scene despite reasonably consistent appearance. Model A closely matches the rainy street, yellow rainco | Anthropic: Claude Fable 5.1: Model A depicts the requested man in a yellow r…
Turbo Tandoor Dash
A 6-second continuous shot in a bustling night-market test kitchen: chef Rafiq Momen, wearing a saffron bandana and charcoal apron, sprints alongside a stainless prep counter while violently whisking a copper bowl of neon-green cilantro chutney, then lunges to slide a tray of skewered paneer into a roaring tandoor; the camera does a low, fast tracking move parallel to him with slight handheld vibration and realistic motion blur on his arms, the bowl, and the passing background, emphasizing explosive speed and momentum; hot furnace light from the tandoor mixes with cool cyan stall lights, sparks and steam streak briefly through frame, and the mood is urgent, athletic, and electrifying, 16:9
OpenAI: GPT-6 Luna: Model A closely follows the action, moving from whisking beside a busy counter to sliding paneer toward a flaming tandoor, with fitting lighting and a cinematic sense of urgency. Model B shows a seated interview in a living room and misses the requested scene and action entirely. (Second judge pass, order swapped — scores are the average of both: Model B shows a composed living-room interview and misses nearly all of the requested action and setting. Model A depicts a fast-moving chef whisking g | Anthropic: Claude Fable 5.1: Model A delivers the prompt closely: a chef in…
Lighting transition
A living room at dusk as the sun sets over a few seconds: the warm light visibly dims and shifts to cool blue while a lamp flickers on, the transition smooth and believable, locked-off frame, 16:9.
OpenAI: GPT-6 Luna: Model A shows the requested locked-off living room transitioning from warm dusk light to cool blue, with the lamp on in the final frame. Model B instead shows a person speaking in a bright room and does not depict the requested lighting transition. (Second judge pass, order swapped — scores are the average of both: Model B shows a person speaking in an evenly lit room, with none of the requested dusk transition. Model A keeps a locked-off living-room view as the warm sunset light gives way to co | Anthropic: Claude Fable 5.1: Model A delivers exactly the requested locked-o…
Matchup powered by OpenRouter.