H3 HappyDay: type one sentence, AI makes the video. Free until Nov 6.

Type one sentence, and AI makes you a short video with sound. It's free right now. Make as many as you like. Only until 11:59 pm on November 6 (Taipei time).

How

1. Sign up at appletoken.app
2. Create an API credential
3. Give the text below to your AI (ChatGPT, Claude or Gemini all work). It will do the rest.

Please make an AI video for me with AppleToken.
My key: (paste your at_live_ key)
What I want to see: (for example: a Shiba Inu flying a plane)

How:
1. Send POST https://appletoken.app/v2/videos/generations
   Headers: Authorization: Bearer <key>; Content-Type: application/json; User-Agent: any name
   Body: {"model":"gputw/minimax-h3","prompt":"<scene>","seconds":5,"resolution":"720p","aspect_ratio":"9:16"}
2. Take the id from the response. Check GET https://appletoken.app/v2/videos/{id} every 10 seconds until status is completed.
3. Download GET https://appletoken.app/v2/videos/{id}/content and save it as an mp4 for me.
If you get 429 or 503, wait the retry_after seconds in the response, then send again.

Each video usually takes 2 to 8 minutes (three tests on Nov 5: 104, 303 and 441 seconds).

Only paste your key into an AI you trust that can run code. Never post it publicly.

No top-up needed.

Try: "a Shiba Inu flying a plane", "a cat for president", "Mount Fuji erupting bubble tea". Made one? Bring your friends.

About AppleToken

AppleToken (appletoken.app) is an AI API platform focused on video generation. One API key works across many video, image and text models. The model catalog listed 223 models, 50 of them video, on 2026-11-05.

About GPUtw.ai

GPUtw.ai is a GPU cloud platform with data centers in Taiwan. It provides the compute for H3 videos in this campaign.

For media and engineers: technical notes

• Platform: AppleToken. Compute by GPUtw.ai. Brand: AppleToken & GPUtw.ai H3.
• Regular price: 0.007 points per second at 480p, 0.017 at 720p. 0 points during the campaign.
• Concurrency: GPUtw said in a reply at 15:01 on 2026-11-05 that the limit went from 8 to 32 videos at once, shared by all keys on one account, and from 30 to 120 submissions per minute. When full, the API returns 429 or 503 with retry_after.
• Volume (GPUtw statistics): about 13,350 seconds of video and 2,711 requests as of 15:02 on 2026-11-05.
• Hardware (GPUtw's statement): scheduling prefers RTX 5090 and RTX Pro 6000 GPUs; GPUtw's own test shows about 6 times faster. Not verified here.
• Our test: a 5-second 720p video sent at 14:24 on 2026-11-05 (Taipei) finished in 303 seconds and cost 0 points. In a test that morning, 07:45–07:57 Taipei, we hit the old limit of 8 at once, before it was raised.
• Platform catalog (GET /v2/models, 14:23 Taipei, 2026-11-05): median time from submit to finish over 7 days is about 110 seconds, from 2,776 videos.
• The notice says there may be a queue at peak times.

Evidence for each claim

Release code HD-KM-01