Space Bunny Alpha in practice: speed and quality of generating games, scenes, and 3D models

In the tests described by the author, Space Bunny Alpha created visual scenes, a game prototype, and a 3D model. The speed and accuracy of the results varied.

These are the author's assessments of individual runs, not the results of a comparative test.

How long Space Bunny Alpha generation took

Some tasks took tens of minutes, while creating a game took almost two hours, according to the author's estimate. They timed some tasks precisely and gave approximate figures for the others.

These are the results of specific runs, not a promise that the next generation will take the same amount of time. The author estimated that the aquarium in XHigh mode would take approximately 40 minutes or an hour. They did not provide a separate precise measurement for this mode.

The game task took particularly long. The author speculated that the model spent a long time thinking through the prompt, but the test does not establish the reason for the delay.

Ice cream: successful details, but imperfect sprinkles

For the ice cream generation, the author noted convincing chocolate drips and a version where chocolate covered the surface in a solid layer. In places, the sprinkles looked like they were floating above the ice cream, so the image was not flawless.

The author rated the image as good and placed it among roughly the five best results they had seen from similar generations. This is their personal assessment, not the result of a comparison using a consistent methodology and predefined criteria.

Aquarium: a lively scene with noticeable errors

The first version of the aquarium appeared in High mode in 19 minutes 8 seconds. The author observed that the rocks and plants moved, bubbles appeared, and the water's surface rippled. Some details looked implausible.

The bubbles crossed the aquarium's boundaries before disappearing. The clownfish also seemed inaccurate to the author. The scene gave an impression of movement, but individual objects and their appearance needed checking.

After a request to improve it, new fish species and colors, crabs, crayfish, and snails appeared. There were more details, but the author rated the updated version lower: the crab's leg movements seemed wrong, and some snails looked strangely designed or positioned.

In this test, adding objects did not improve the scene. If realistic animation and careful composition matter, each element should be checked, not just the overall impression.

Game: many mechanics, repetitive gameplay

Creating the game in High mode took almost two hours, according to the author's estimate. The prototype featured shooting, grenades, a tank, changing locations—including a nighttime setting—people to rescue, and a boss. The author considered the result substantial for a rough demonstration of a range of ideas.

The set of mechanics did not make for varied gameplay. The author noticed that enemies appeared mostly from one side, so the player could simply keep shooting in that direction. There were hardly any hiding places or vertical movement, such as moving up and down.

As a result, the world felt flat, and the gameplay loop quickly became repetitive. The author described the graphics as pixelated and rated them as inferior to the results from other models they had tested. According to the author's observation, the tank that was created could only fire its machine gun and had no other functions.

The author also compared the duration with another personal experience: they said GPT created a similar game in about 15–18 minutes. This is a personal comparison; the account does not include data that would allow the prompts and settings to be compared.

3D model from an image: limited resemblance to the source

Building a 3D model from an image took 13 minutes 49 seconds. In the author's assessment, the result did not accurately capture important features of the original character.

The model did not preserve the hands in the pockets, slippers, or sword, and the hat did not match the prompt. The author considered it to bear little resemblance to the source image. In this test, the specified pose and accessories were reproduced inaccurately.

Who Space Bunny Alpha is suitable for

Based on these tests, Space Bunny Alpha can be used for rough visual experiments if the user is prepared to wait and manually check the result. For the ice cream generation, the author noted convincing materials and effects, while the aquarium produced a moving scene with several types of objects.

The model seemed less convenient to the author for rapid game prototyping: in the described test, the wait approached two hours, and the gameplay, in their assessment, would have required substantial reworking. In the 3D model test, the specified pose and accessories were reproduced inaccurately.

If a predictable turnaround time, an accurate resemblance, or a ready-to-play game balance matters, a single run like this is not enough. You can first test the task with a small prompt, record the time, and assess how much correction the result will need after generation.

What is known about the model's origins

The author speculated that Space Bunny Alpha might be based on MiniMax M3.5 or M4, and also suggested it could be another little-known Chinese model. These are speculations: according to the author, only MiniMax's developer knows the exact version.

These guesses should not be treated as established facts or used as a basis for comparison with specific products. The narrower conclusion supported by the described runs is that Space Bunny Alpha handled various tasks, and the author rated it as average in quality and slow.

Final assessment of Space Bunny Alpha

In the tests, Space Bunny Alpha produced some successful details in visual generation, but the author considered the complex scenes and game prototype to require careful scrutiny. The aquarium took 19 minutes 8 seconds to generate, while the 3D model took 13 minutes 49 seconds. The game took almost two hours, according to the author's estimate.

For a one-off experiment where getting a starting point matters more than a finished product, this wait may be acceptable. The author's experience with rapid iterations, accurately reproducing a reference, and developing a game with varied gameplay is reason to compare alternatives first using identical prompts.