We have a new top AI video generator!

Key Concepts

• Hyo 2.3: The latest version of the AI video generator Hyo, with significant improvements in physics, high action shots, and world understanding.
• Sora 2 & V3 3.1: Other leading AI video generation models used for comparison.
• Text-to-Video Generation: Creating videos solely from textual descriptions.
• Image-to-Video Generation: Creating videos by using an image as a starting point.
• Physics Understanding: The AI's ability to accurately depict physical phenomena and interactions.
• World Understanding: The AI's comprehension of real-world objects, characters, and their relationships.
• Prompt Engineering: Crafting detailed and specific text prompts to guide AI generation.
• Anatomical Accuracy: The AI's ability to generate human and animal figures without deformities or inconsistencies.
• Cinematic Movements: The use of dynamic camera angles and character actions to create visually engaging scenes.
• Guardrails: Restrictions or limitations imposed on AI models to prevent the generation of certain content.
• Presets & Camera Control: Features within Hyo 2.3 that offer predefined prompts and camera movement options.

Hyo 2.3: A New State-of-the-Art AI Video Generator

This video introduces Hyo 2.3, the latest iteration of the Hyo AI video generator, highlighting it as a significant upgrade from its predecessor, Hyo 2. The presenter emphasizes its strengths in generating high action shots, demonstrating accurate physics, and exhibiting strong world understanding. The review includes comparisons with other leading models, Sora 2 and V3 3.1, to showcase Hyo 2.3's capabilities and limitations.

High Action Shots and Cinematic Movements

Hyo 2.3 excels at generating dynamic and high-action scenes.

• Example: A prompt describing a sorceress casting fireballs while an opponent summons icy dragons, with their powers clashing mid-air and dynamic camera pans, resulted in an "epic and higher action" video.
• Comparison: In contrast, Sora 2's generation for the same prompt was described as being in "slow mode," and V3 3.1's movements were also slow and lacked the epic feel.
• Observation: While Hyo 2.3's generation was impressive, some "noise and distortions" were noted around the edges, particularly on the icy dragon.

Physics and World Understanding

The presenter tested Hyo 2.3's ability to accurately depict physical phenomena and complex scenarios.

• Unicycle Juggling:
  • Prompt: "A man riding a unicycle and juggling red balls."
  • Hyo 2.3 Result: Successfully generated a man juggling red balls while riding a unicycle. The juggling was "pretty good," though the unicycle's balancing aspect was less dynamic than expected, appearing "fixed in that position."
  • Comparison: Both Sora 2 and V3 3.1 "completely fail at juggling." Sora 2's character was "just throwing balls everywhere," and V3 3.1's attempt was deemed "even worse." Hyo 2.3 was awarded the point for this prompt despite not being perfect.
• Freezing Water:
  • Prompt: "A time lapse of water in a glass that is left outside in the cold where the water slowly freezes."
  • Hyo 2.3 Result: The generation was "pretty good," though "exaggerated" with the water level rising "too much." The presenter noted that real-life ice is less dense than water, causing the water level to rise slightly upon freezing.
  • Comparison: Sora 2 "never really turns into ice" and its water level rise was "way too much." V3 3.1's generation was "completely wrong." Hyo 2.3 was considered the most accurate among the three, despite not being entirely correct.
• Complex Scene Understanding:
  • Prompt: "A ballerina in a tutu practicing spins in a studio with mirrored walls scattered with point shoes and sheet music. A rabbit watches on top a grand piano. Outside, an elephant balances on a circus ball."
  • Hyo 2.3 Result: Successfully generated most elements: a spinning ballerina (anatomically correct), a studio with mirrored walls, scattered point shoes and sheet music. A rabbit was present on the piano bench (not the piano itself). An elephant was balancing on a circus ball outside the window. The ballerina's spins were "very accurately" generated.
  • Comparison: Sora 2 was "not bad," with the elephant balancing, but the ballerina wasn't truly spinning. V3 3.1 added "three rabbits to the piano," the elephant was "completely wrong" and not outside the window, and the ballerina's spin had anatomical errors ("her front kind of switches with her back halfway"). Hyo 2.3 was again favored for getting most elements correct.

Handling Multiple Characters and Actions

Hyo 2.3 demonstrates proficiency in generating scenes with numerous characters and complex interactions.

• Ninja Ambush:
  • Prompt: "A group of ninjas ambushing a heavily armored samurai in a bamboo forest with sword strikes, acrobatic flips, and leaves swirling in the wind."
  • Hyo 2.3 Result: Generated a lone samurai in a bamboo forest, followed by an army of ninjas attacking him. The scene was "pretty good" with "cinematic" camera movements and character actions, though some "distortion and noise" were present on edges of characters and swords.
  • Comparison: Sora 2 generated the scene but was again in "slow motion." V3 3.1 "cannot generate fight scenes very well."

Emotional Range and Realistic Portrayals

The model can generate videos depicting a range of human emotions and realistic character appearances.

• Emotional Transitions:
  • Prompt: "A young woman laughing very hard. Then, she looks shocked. Then, she bursts out crying, then she looks really excited."
  • Hyo 2.3 Result: Successfully followed the prompt, showing the woman laughing, looking shocked, crying, and then excited. The generation was described as "pretty good" and "realistic," with normal-looking teeth, avoiding the "perfect polished and plasticky face" seen in older models.
  • Comparison: Sora 2 and V3 3.1 were also "very good" in this instance, resulting in a tie.
• Figure Skating:
  • Prompt: "A young figure skater gracefully ice skating on a frozen river that winds through a snowy mountainous canyon. The camera follows her dynamic movements as she skates and twirls. It's going to be a fast tracking shot."
  • Hyo 2.3 Result: Generated the specified scene with accurate spinning and no deformed limbs. The generation was "very cinematic" with good tracking camera movements.
  • Comparison: Sora 2 had "nice motion and camera effects" but the skater "flies off into the horizon" at the end. V3 3.1 had "anatomical errors," with legs switching. Hyo 2.3 was preferred for its accuracy and cinematic quality.

Generating Existing Characters (Text-to-Video)

A notable feature of Hyo 2.3 is its ability to generate videos of existing characters from text prompts, a capability not offered by all competitors.

• Will Smith Eating Spaghetti:
  • Prompt: "Will Smith eating spaghetti."
  • Hyo 2.3 Result: Successfully generated Will Smith eating spaghetti, noting the "huge plate" and the character's "depressed" look. The presenter stated, "Hyo is the only commercial video model that actually allows you to generate celebrities and existing characters."
  • Comparison: Sora 2 "won't let me generate this." V3 3.1 generated "someone, but this is not the Will Smith that I was going for." Hyo 2.3 was the clear winner in this category.

Image-to-Video Generation

Hyo 2.3 also supports image-to-video generation, using an uploaded image as the starting frame.

• Soldiers vs. Tentacle Monster:
  • Input: Image of a chaotic battle scene between soldiers and a tentacle monster.
  • Prompt: "An epic fight scene of soldiers versus a giant tentacle monster in the desert, high action, motion blur, intense cinematic, shaky camera. First-person view of the soldier."
  • Hyo 2.3 Result: Generated an "epic scene" where a soldier reloads ammo. Some "noise" was present, especially with background elements, which was deemed "expected" for such a high-action scene.
  • Comparison: Sora 2's generation showed characters "frozen there," not moving. V3 3.1 changed the tentacle monster's appearance (adding a mouth) and had "slow motion" movements despite the prompt. Hyo 2.3 was again favored.
• Warrior vs. Monster (Fly-through Shot):
  • Input: Image of a warrior and a monster.
  • Prompt: "The warrior to sprint towards the monster. Then he leaps towards the monster getting ready to strike. The monster opens its mouth and breathes fire engulfing the warrior in flames. It's going to be an epic fly-through shot tracking the warrior closely."
  • Hyo 2.3 Result: The warrior sprinted, leaped, and was engulfed in flames as the monster breathed fire. The prompt was "nailed," with the only error being an inconsistent sword. The scene was "very good and cinematic."
  • Comparison: Sora 2's warrior ran "so slowly." V3 3.1's movements were "way slower," and the warrior's final action was unclear. Hyo 2.3 was superior.
• Anime Scene:
  • Input: An anime scene with characters and motorcycles on a highway.
  • Prompt: Left empty.
  • Hyo 2.3 Result: Generated a "pretty good" scene resembling an anime show, retaining details of characters and motorcycles with "nice motion."
  • Comparison: Sora 2 failed to generate them "actually driving through this highway," remaining "still the whole time." V3 3.1 flagged the content as "sensitive" and refused to generate a video. Hyo 2.3 was the only successful generator.
• Busy Marketplace:
  • Input: A busy photo of a marketplace with many stalls, items, and people.
  • Prompt: Left blank.
  • Hyo 2.3 Result: "Not bad," but with "some warping on some of the faces," especially in the background.
  • Comparison: Sora 2 "would not allow me to upload any photos of realistic people." V3 3.1 also had "a lot of warping and noise" in the background. Both Hyo 2.3 and V3 3.1 exhibited similar issues with busy scenes.

Limitations and Areas for Improvement

Despite its strengths, Hyo 2.3 has certain limitations.

• Text Generation:
  • Prompt: "A professor explaining the Pythagorean theorem on the whiteboard."
  • Hyo 2.3 Result: Generated text that was "not really good" and did not accurately depict the Pythagorean theorem ("Pyagga nana nana theorem").
  • Comparison: Sora 2 was "the closest," writing the formula at the top but with incorrect diagrams. V3 3.1 was "just not correct." None of the models could accurately generate this prompt.
• Technical Concepts:
  • Prompt: "An instructional motion graphic video showing how data flows through an artificial neural network."
  • Hyo 2.3 Result: "Not correct," not resembling data flow in an ANN.
  • Comparison: Sora 2 was "the closest," resembling an ANN's appearance. V3 3.1 was "completely wrong." Again, none were fully accurate, but Sora 2 was closer.
• Audio: Hyo 2.3 does not have built-in audio generation, requiring integration with other tools for sound.
• Resolution and Duration: Currently supports 768p or 1080p resolution, with 6-second or 10-second durations. 1080p generation is limited to 6 seconds.
• End Frames: Does not support uploading an image as an end frame in version 2.3; this feature is available in Hyo 02.

Features and User Experience

Hyo 2.3 offers features to enhance user control and creativity.

• Presets: Provides predefined prompts to guide generation.
• Camera Control: Allows users to select specific camera movements (e.g., orbit, tilt, tracking shots) to add cinematic effects.
• Free Trials: Offers four free trials per day, encouraging users to experiment with the platform.

Conclusion and Comparison Summary

Hyo 2.3 is a powerful AI video generator, particularly strong in:

• High action, high movement scenes: Excels in fight scenes and battle sequences.
• Complex prompts: Handles prompts with numerous elements effectively.
• Physically challenging prompts: Demonstrates good performance in scenarios like figure skating or ballet.
• World understanding: Accurately generates existing characters and complex scenarios.
• Fewer guardrails: Offers more creative freedom compared to some competitors.

It surpasses Sora 2 and V3 3.1 in these areas, provided that audio generation is not a primary requirement. Its limitations lie in text generation, accurate depiction of complex technical concepts, and the absence of built-in audio. The presenter recommends taking advantage of the free trials to explore its capabilities.