Wavespeed
English
Start creating

Practical guide

How to approach the wavespeed playground

Treat the wavespeed playground as a place to test an idea, not as a guarantee that every model accepts the same inputs. Start with one specific result you want, then change one part of your request at a time.

Opens a separate AI platform
Creative visual displayed in the Wavespeed site hero

Prerequisites

A small amount of preparation makes the first result easier to judge and the next attempt easier to improve.

Visual example from the site's feature imagery
01

Bring a brief you can actually evaluate

Write down the subject, setting, visual style, and one detail that must survive the generation. For example, a product concept might need a matte ceramic cup, a wooden surface, soft morning light, and no visible lettering. If the result misses the cup's material, you know what to revise; if the request merely says “make a nice image,” you do not.

  • Keep the first request short enough to inspect.
  • Separate essential details from optional decoration.
Second visual example from the site's feature imagery
02

Check the available input before preparing assets

A prompt is useful for a text-based test. A reference image is useful only when the selected tool accepts one and you have permission to use it. Check the chosen model's visible input requirements, supported files, and output controls before uploading anything. Save your original prompt outside the workspace so you can compare attempts even if you switch tools.

  • Use material you own or are allowed to upload.
  • Record which tool and settings produced each result.

One full run-through

Use the same brief twice. The second pass should address an observed problem, rather than replacing the whole idea.

First pass
Focused revision

Goal

First pass

Choose a single image concept: a ceramic espresso cup on a walnut table.

Focused revision

Keep that concept; decide that the cup's material needs improvement.

Input

First pass

Enter the subject, surface, lighting, depth of field, and no-lettering request.

Focused revision

Reuse the prompt and clarify “matte, unglazed ceramic” instead of adding a new scene.

Tool check

First pass

Confirm that the selected option accepts a text prompt.

Focused revision

Keep the same option where possible so the comparison remains meaningful.

Submission

First pass

Review the visible settings, then submit the request.

Focused revision

Change only the material wording and submit again.

Inspection

First pass

Check the subject, surface, light, and unwanted lettering separately.

Focused revision

Check whether the material improved without losing the other details.

Record

First pass

Save the prompt and note what worked and failed.

Focused revision

Save the revised prompt beside the first result and choose the stronger version.

Options table

Match the kind of test to the inputs the selected tool actually offers. These are planning choices, not a promise of particular controls.

Best for a fresh concept

Start here when you have an idea but no source asset. Describe the main subject before adding lighting or style. If the result is too generic, replace one vague adjective with a visible detail; “side-lit by a window” is easier to assess than “beautiful.”

Input to prepare: a compact prompt.
First check: whether the main subject appears as requested.

Best for retaining a visual detail

If the chosen tool accepts an image, use a reference when shape, composition, or an existing object matters more than a wholly new scene. Inspect the upload requirements first. Avoid sharing private photographs or material you cannot authorize for use.

Input to prepare: an appropriately licensed image.
First check: which reference details remain recognizable.

Best for testing an action

If a video-capable option is available, describe one action and a clear starting scene rather than several competing events. Judge the resulting motion separately from the appearance of individual frames. A still-image result cannot establish how a movement instruction would perform.

Input to prepare: one subject and one action.
First check: whether the intended action is legible.

What fails

Make the next attempt easier to diagnose

Crowded prompts make it hard to tell which instruction caused a poor result. Switching tools, references, and wording all at once makes comparisons unreliable. An unsupported file or input type can stop a test before generation begins. When a result disappoints, keep the useful parts of the brief, name the most important failure, and revise only that part. You can take the example prompt to a separate AI platform to begin a fresh test.

Try a focused prompt
  • Check the selected tool's inputs before submitting.
  • Compare results against the same written goal.
  • Change one meaningful detail per revision.

Playground FAQ

Choose a simple subject with a few observable details, such as an object, a surface, and a lighting condition. A narrow request gives you something concrete to evaluate before you attempt a more complex scene.

That depends on the input requirements of the option you select. If it accepts text alone, a written brief is enough for an initial test; prepare an image only when an available workflow calls for one.

Model choice, settings, input assets, and prompt wording can all affect the output. Record those details and change one variable at a time if you want a useful comparison.

Check whether the selected option supports the input you supplied, then identify the single most important mismatch. Rewrite that instruction in concrete visual terms and try again without changing the rest of the brief.

Start creating
Start creating