Gemini 3.8 Flash: SVG Animation Generation Case — Ninja Riding a Flying Dragon (X)

One-sentence takeaway

The author used a one-line prompt to have Gemini 3.8 Flash generate an SVG animation of “a ninja riding a bicycle on a flying dragon,” and attached the GIF result to the original post. The author said this was the first SVG test and that the result was better than GPT 5.6 Sol, but the post provided no independent score, source code, or reproducible parameters.

User template

The following is a replaceable template organized from the original post's phrasing; it is not the original post's exact text:

Create a self-contained SVG animation of a [original subject] [action] [scene or carrier object].

For more stable reproduction, you can add constraints to the template such as a viewBox, animation duration, loop behavior, whether external resources are allowed, and “output a single SVG file that can be opened directly in a browser.” The original post did not provide these constraints.

Original post prompt

The only prompt explicitly made public in the original post is shown below, with its original capitalization and wording preserved:

create a SVG Animation of a Ninja Riding a Bicycle on top of a Flying Dragon.

Chinese meaning: Generate an SVG animation of a ninja riding a bicycle on the back of a flying dragon.

Input and result

Input

• Model: Gemini 3.8 Flash (as attributed by the author for this test)

• Prompt length: One-sentence English natural-language description

• Task: Generate an SVG animation

• Unspecified: Dimensions, viewBox, frame rate, duration, loop behavior, SVG version, external images, source filename, and runtime environment were not stated

Original post result

• The original post displayed an embedded GIF as the result of this SVG test.

• The author said Gemini 3.8 Flash performed better than GPT 5.6 Sol in this first SVG test.

• The original post did not publish the SVG source code, an online preview URL, frame-by-frame notes, failure examples, or an objective score. Therefore, it only confirms that “a displayable animation result” existed; it cannot reproduce the same image or prove that the model is generally stronger.

Reusable approach

1. First, use the one-line prompt above to generate an SVG animation with a fully original subject.

2. Open the result in a browser and confirm that the SVG actually contains animation rather than only a static image.

3. If the output will be delivered or reused, add constraints: single file, self-contained, fixed viewBox, no external resources, specified duration, and looping.

4. To compare models, keep the same prompt, runtime environment, and acceptance criteria, and record generation time, whether the source runs, animation completeness, and visual defects; the original post did not apply these controls.

Scope and limitations

• This is a one-off SVG generation demonstration, not a systematic benchmark; “better than GPT 5.6 Sol” is a self-reported result from the author.

• The one-line prompt did not specify output format or animation constraints. With a different model, client, or renderer, it may produce a static SVG, non-runnable code, or a different composition.

• Ninjas, flying dragons, and bicycles are generic creative elements. When reusing the approach, do not replace them with specific copyrighted characters, brand logos, or real people, and do not publish the original post's GIF as your own material.

• The original post did not disclose the API model ID, reasoning level, temperature, token limit, generation time, or cost; these parameters cannot be inferred from the case.

Source observations

This post was a reply under Philipp Schmid's post about the Gemini 3.8 Flash Cursor benchmark. Suyash's reply provided both a copyable prompt and a GIF result, making it a useful real-world example of “short prompt → SVG animation.” It is unrelated to the YouTube video analysis and editing-reuse workflows already in this directory.