PromptLab.

Original: Outside a convenience store on a rainy night, a delivery rider stops, removes their helmet and checks the last order on their phone.

Refinement turns the scene into model-oriented camera and action instructions, such as framing, camera movement and action order. This is an illustration; the selected model produces the actual result and may add lighting, sound or other details.

Choose Seedance 2.5 for a Seedance video prompt, or MiniMax H3 for a MiniMax video prompt. Switching models does not submit a request or use your allowance. MiniMax uses its native IR service; Seedance uses the existing prompt optimization rules.

Results keep the model’s native format and may use a different language. MiniMax IR may return structured English descriptions. Both refiners may add camera, lighting, sound or other details. Review the complete result before use. Your original text remains available for editing.

Guests get 2 per day; members get 3; Spark gets 10; Studio gets 20; Director gets 50. Both models share the allowance. It resets at 00:00 UTC (08:00 Taipei). Signing in merges today’s guest usage: a guest who has used 2 refinements has 1 left after signing in as a standard member.

After the free allowance, members confirm 1 credit for each request. Credits are consumed only after the complete result succeeds. Cancelling the confirmation submits nothing; a confirmed failure releases the reservation. An uncertain result keeps its reservation while the same request is checked. Signing in or returning from credit purchase preserves your draft and model; you submit it yourself.

Prompt Lab refines text only. It does not upload media, generate images or videos, or create projects. Copy the complete prompt when it is ready. The next-step button opens the existing storyboard image composer, where you prepare and submit a separate request.