Loy Krathong 24 Nov 2026: a guesthouse clip with eco-krathong cues

Loy Krathong 2026 is Tuesday 24 November, with Chiang Mai's Yi Peng lantern days on 23 to 25 November, according to one travel guide. A guesthouse, hostel or tour desk can turn one riverside photo into a short silent clip and burn the date and a materials reminder onto it as caption cues.

The dates and the materials advice come from the Visit Thailand Today guide, read on 2026-10-13. It is a travel guide, not an official notice, so check local announcements before you promise an event. Everything about Sume comes from the docs pages under sources.

What does the guide say about dates and krathong materials?

The guide gives Loy Krathong as Tuesday 24 November 2026, on the full moon of the twelfth month of the Thai lunar calendar, and Chiang Mai's lantern festival as 23 to 25 November. It discourages foam krathongs, says they are banned in Bangkok's parks, and calls a banana-trunk and banana-leaf krathong the best choice. It also lists park rules banning foam krathongs, sky lanterns, fireworks, firecrackers and alcohol sales.

A clip for guests should repeat the reminders that apply where you are. Park rules in Bangkok are not rules in another province.

Which Sume calls build the clip?

Step one is a video job at POST /v1/videos. Put your photo in frame_images with frame_type set to first_frame and Sume treats the job as image-to-video; send no photo and it is text-to-video. Before you pick a model, call GET /v1/videos/models and read supported_durations, supported_frame_images and supported_aspect_ratios for it, because limits differ by model. The docs state, for example, that wan-3.0 accepts 2 to 30 seconds.

Step two is POST /v1/video-captions with the finished clip's public HTTPS URL as video_url and a cues array of text, start and end. Because you send cues, Sume does not run speech-to-text and burns your text at those times. A silent clip with no cues fails with caption_no_speech, so the cues are not optional for a silent clip. Both calls are asynchronous jobs; poll or take a webhook for the result.

What should the cues say?

Put the date first, then one practical line per cue. A five-cue plan for a 10-second clip:

How do you send the caption job?

This script builds the cue payload and prints it, so you can check the JSON before you post it with your own API key and the real clip URL.

import json

cues = [
    (0, 2, "Loy Krathong is Tuesday 24 November"),
    (2, 4, "Choose a banana-leaf krathong"),
    (4, 6, "Please skip foam"),
    (6, 8, "Ask us for the nearest pier"),
    (8, 10, "Happy Loy Krathong")
]
payload = {
    "video_url": "https://example.com/loy-krathong.mp4",
    "cues": [{"text": t, "start": s, "end": e} for s, e, t in cues],
}
print(json.dumps(payload, indent=2))

What can go wrong?

Host the clip at a public HTTPS URL first. A riverside photo with floating lights is a hard still for generated motion: lights can multiply or drift, so look at the result before you caption it. A calm, short clip is safer than a long one.

Do not caption a promise you cannot keep. If your pier is closed on the night, say so in the post text and keep the cues to the date and the materials reminder.