Author: Changan I Biteye Content Team

At its launch event, Google showcased the capabilities of Gemini Omni, which can generate videos from text, images, audio, and video as input, and allows for continued editing through conversation. The official statement says it will replace Veo in the Gemini App, supporting 10-second videos, native audio generation, reference image generation, and video-to-video editing.

Prior to this, creators discussing AI videos could not avoid Seedance 2.0, whose advantages were clear: stable motion, strong camera presence, joint audio-visual generation, and good overall short video quality.

So here's the question:

What are the differences between Gemini Omni and Seedance 2.0? Which one is more suitable for creators? This article compares them from three practical perspectives: price and cost, product experience, and video quality.

I. Price Comparison: The real expense in video models lies in the "cost of discarded footage".

Let's start with the most practical question: How much does it cost per use?

When comparing Gemini Omni and Seedance, pricing shouldn't be based solely on subscription fees; two issues need to be considered:

What is the cost of generating a single video?
How many times do I need to retries to get a usable video?

1. Gemini Omni: More like a "subscription + Flow Credits" model

Gemini Omni uses a combination of Google AI membership and Flow Credits.

Based on the AI Ultra premium trim, 25,000 credits per month can generate approximately 8,333 seconds of video. From this perspective, Gemini's pricing structure is clear: the higher the trim level, the more significant the decrease in cost per second.

The cost of AI Plus is approximately 0.82 yuan per second, which isn't particularly low; however, with AI Pro, the cost drops to 0.41 yuan per second. For the Ultra tier, as long as the monthly allowance is fully utilized, the cost per second can be reduced to around 0.2 yuan per second.

Therefore, the Gemini Omni is more suitable for two types of people:

For existing Google AI Pro/Ultra users, the video capabilities are essentially integrated directly into their existing subscription system.
For high-frequency video creators, the higher the output, the lower the average cost.

⚠️However, please note that this cost is calculated based on US pricing. For Chinese users, actual usage will depend on account region, payment method, and access stability.

2. Seedance: Based on the original annual fee, 140 points are consumed every 10 seconds.

Jimeng uses a membership + points system: members receive a certain number of points each month, and points are deducted when generating videos based on the model, duration, and resolution.

Based on the premium tier, 6160 points per month can generate approximately 440 seconds of video.

From this perspective, Seedance's cost is roughly stable at around 1 yuan per second. Basic membership costs approximately 1.06 yuan per second, standard membership approximately 1.00 yuan per second, and premium membership approximately 0.98 yuan per second.

Unlike Gemini, the cost per second difference between Seedance tiers is not significant. While the premium membership is cheaper, its main advantage is not a substantial reduction in cost per second, but rather the increased video length that can be generated each month.

Its advantages are more geared towards the domestic product experience: the entry point is more direct, the payment is more convenient, the Chinese environment is more user-friendly, and it is more suitable for domestic creators to get started quickly .

II. Product Experience: Gemini is more like a workflow, while Seedance is more like a creation tool.

1. Generation efficiency: Gemini is faster, Seedance has a longer waiting time.

Let's look at the generation speed first.

In my actual testing, Gemini Omni takes about 2 minutes to generate a video, while Seedance takes longer, usually about 5-6 minutes.

Gemini's advantage lies in its faster trial-and-error speed. You can see results more quickly and maintain a creative state more easily.

The problem with Seedance is that each wait time is longer, and if the result is not satisfactory, a new version needs to be generated, which quickly increases the time cost.

Therefore, in terms of efficiency, Gemini is more suitable for quick prototyping and inspiration testing, while Seedance is more suitable for final generation after the image has been clearly conceived.

2. Candidate Version: Gemini can generate multiple versions at once, meaning Dream leans more towards single-generation.

Gemini Omni offers x1 / x2 / x3 / x4 options. For example, a 10-second video costs 30 credits, but if you choose x2, it will cost 60 credits.

Because AI videos have a high degree of uncertainty, the effects generated by the same prompt can vary greatly. By generating four versions at once, users can directly choose the best one instead of re-generating them one by one.

The dream-generating aspect, on the other hand, leans more towards single-time generation.

Therefore, we can conclude the following:

Gemini is better suited for quick card drawing and selection.
Immediately, it is more suitable for single-line production after the parameters are clearly defined.

3. Video Length: Maximum 15 seconds for JiMeng, maximum 10 seconds for Gemini.

The third difference is the video length.

Based on current product experience, Gemini Omni can generate videos up to 10 seconds long, while Jimeng can generate videos up to 15 seconds long.

If you want to create a slightly more complete shot, such as a person entering the room and sitting down, a product from a long shot to a close-up, or a short dramatic action, 10 seconds may not be enough.

Although 15 seconds is not long, the extra 5 seconds compared to 10 seconds is enough to complete an additional action or a shot change in the video.

Therefore, in terms of duration:

Gemini is better suited for short shots and quick footage.
Seedance is more suitable for slightly more complete video clips.

If you're creating short dramas, commercial storyboards, or emotional narratives, Seedance's 15-second limit will be more useful.

4. Video Editing: Gemini can be further modified, Seedance is more like a regeneration.

In video generation, a very real problem is: what if you're not satisfied with the first version?

Image generation is manageable; re-generating an image isn't too costly. However, once video generation involves duration, queuing, points, and review, each restart increases the cost.

From a product experience perspective, Gemini Omni places greater emphasis on "editing" capabilities. It doesn't just allow users to rewrite the prompt and regenerate; instead, it places the video into a more continuous creative workflow: first, a version is generated, and then modifications are made based on the existing results, such as adjusting the visuals, changing the style, replacing some elements, or further optimizing the shots.

Gemini's advantage lies in the fact that it makes video generation more like a "continuously communicative" process.

Seedance currently feels more like a traditional image generation tool. Its strength lies in the generated results themselves, especially the visuals, motion, and overall cinematic quality. However, if you're not satisfied with the first version, you'll often need to readjust the prompts, reference images, and parameters to generate a new version.

5. Review Mechanism: Different Restrictions on Both Sides

Review is an unavoidable issue in actual use, and from my testing, the focus of the review on both sides is not quite the same.

Gemini Omni's review process for video content is generally quite strict. One notable example is that some content, even when presented in a chibi or cartoonish style, may still fail to pass review. In other words, a more "fictional" art style doesn't automatically lead to more lenient restrictions.

Interestingly, Gemini isn't as strict as I expected when it comes to some live-action content. As long as it doesn't clearly involve sensitive figures, copyright infringement, misleading information, or high-risk content, some live-action style videos can still be generated.

Seedance's review process leans more towards the risks associated with real people and their likenesses. Restrictions are more likely to be triggered when content involves real people, celebrity look-alikes, similar images of famous people, film or television characters, or public figures. Even if the creator simply intends to create an entertaining expression, it may still be blocked.

III. Video Quality: The real difference lies not in individual frames, but in the quality of the video when it's in motion.

At the video level, the most critical question becomes: can the generated video actually be used?

This time, I didn't conduct particularly complex tests. Instead, I started with the scenarios most commonly used by creators and created several sets of actual data. These included aspects such as character combat, generating videos from reference images, and character consistency.

The most obvious feeling right now is:

Gemini Omni has strong video capabilities, but its expression style is more suited to overseas contexts; Seedance will be more natural in the visual style, character movements and anime context that Chinese creators are familiar with.

1. Content comprehension: Gemini allows for more freedom of expression, but tends to lean towards an English context.

I tested a fairly typical video scenario:

Generate a video of two characters fighting.

This test may seem simple, but it is actually very suitable for observing the model's ability to "play freely".

Because I didn't strictly limit what the characters said, nor did I explicitly specify the language of the dialogue, Gemini Omni became more proactive in completing the video content and even automatically generating dialogue and audio.

But here's the problem: Gemini's free interpretation is clearly more biased towards overseas contexts.

Without language restrictions, the characters in the generated videos speak English. This indicates that Gemini's default expression is closer to the logic of English content production. If you want to create short Chinese videos, Chinese fan-made content, or narrative content within a Chinese context, you need to explicitly restrict the language in the prompt.

2. Character Consistency: Gemini has the ability to reference images, but there will still be deviations in the characters.

I also tested generating videos from reference images.

This test mainly examines one question:

Once a reference image of a person is provided, can the people in the video still maintain the same appearance?

As a result, although Gemini Omni can generate videos based on reference images, the characters in the final video will still differ from the original reference images, and it cannot completely maintain the consistency of the characters.

The details of the figures may differ slightly from the references.

If it's just for creating an atmosphere video, a funny video, or a creative short film, slight changes in the characters are acceptable. But if it's for a fixed IP character, Omni is completely inadequate.

3. Continuity of movement: Gemini sometimes feels like a series of fragmented movements, while Seedance feels more like a complete action.

Another obvious problem in the fight video is that the characters' movements are not fluid enough.

In the fight videos generated by Gemini, the transitions between character movements are rather stiff, and it feels like the action wasn't filmed continuously, but rather several action clips pieced together.

For example, a character's movements, from preparing to attacking, dodging, and moving, are all present, but the transitions are not natural enough. This makes the video look like "AI stitching together several keyframes" rather than a real, continuous motion.

This is more noticeable in scenes involving fighting, running, jumping, turning, or multi-person interaction. The problem is less pronounced in simple camera movements, slight character swaying, or product demonstrations.

In these kinds of scenarios, seedance is closer to the creator's expectations. Its movements and shots are usually more like a complete video clip, especially in anime, fighting, and short video action scenes, where the overall rhythm is more natural.

4. Video Editing: Gemini's ability to edit and modify videos is one of its biggest advantages.

However, Gemini also has a very important advantage at the video level: it supports AI modification of existing videos.

In other words, you can upload a real-world video and let Gemini edit the content using AI. This includes modifying visual elements, replacing parts of the content, changing the style, or combining real-world footage with AI-generated content.

This is different from Seedance's product boundaries.

If you want to combine "real-world footage + AI footage", Seedance's current common practice is to rely more on transitions: first shoot a real-world video, then generate an AI video, and use editing transitions to connect them.

IV. In conclusion: Seedance generates more content, while Omni is more like a video editor.

The biggest takeaway from this comparison is that Seedance and Gemini Omni have different focuses.

Seedance is more like a mature AI video generation tool; its core capability is to turn text or images into videos.

In the two most common scenarios of text-based videos and image-based videos, Seedance is still more stable. Its visuals are closer to the aesthetics of Chinese creators, the movements are more fluid, and it is easier to create videos in scenarios such as anime, fighting, and short videos.

If your needs are to turn a prompt into a video or an image into a dynamic video, then Seedance is currently the best choice.

But Omni's highlight is that it supports AI-powered modifications to existing videos.

This means that users don't necessarily have to start with a blank prompt, but can continue to process existing materials: change the background, change the style, replace elements, and even merge real-world footage with AI visuals.

If you're generating video, Seedance is now even more powerful.
If you're looking to edit videos, Omni has more potential.

The next stage of AI video may not just be "generating a video," but rather enabling videos to be directly modified by AI, just like images.

From this perspective, Seedance represents a more mature generation capability, while Omni is more like exploring the future of video editing workflows.

OpenAI founding member Andrej Karpathy joins large model company Anthropic

Popular Articles