# AI Dunk — AI Dunk prompt guide

Canonical HTML page: [https://aidunk.net/ai-dunk-prompt](https://aidunk.net/ai-dunk-prompt)
Last substantive review: 2026-10-27 (Asia/Shanghai).
Purpose: Adapt a photo-to-video basketball prompt and review one controlled APIMart Wan 3.0 result.

The AI Dunk prompt guide is for a manual image-to-video workflow outside the guided AI Dunk preset creator. Attach an authorized appearance image, describe one continuous vertical basketball shot, and specify the run-up, gather, jump, rim finish, landing, framing, visible ball and rim, and unwanted text or extra people.

One controlled test completed on October 27, 2026 through AI Dunk's APIMart adapter. It used an AI-generated fictional adult chef image, the exact 8-second prompt, APIMart Wan 3.0 model wan3.0-video, photo plus text, no reference video, 9:16, 480p, and model-generated audio off. The actual output is 8 seconds, 480×854, H.264, and has no audio track. Manual review found one stable chef, a continuous run-up, gather, jump, one-handed rim finish and landing, with no extra person, duplicate ball, visible brand, caption or watermark. APIMart reported 2.6304 provider credits, or $0.26304, for that run; these are not AI Dunk product credits. This one result does not prove repeatability, a success rate, fixed generation time, or cross-model compatibility. The 6-second and reference-video versions remain untested.

Possible input adjustments include reducing turns or occlusion for face drift, simplifying clothing descriptions, requesting one continuous shot, reducing simultaneous hand and ball actions, keeping the rim visible, avoiding real team or venue branding, and returning to one person when a model does not clearly support multiple references. These are changes to try, not guaranteed fixes.

Read and copy the [full AI Dunk prompt](https://aidunk.net/ai-dunk-prompt), compare the [guided tutorial](https://aidunk.net/ai-dunk-video-tutorial), or use the [guided generator without a written prompt](https://aidunk.net/#dunk-builder).
