Douyin Live Stream UI Prompts

September 15, 2026

Ask an image model for "a typical Douyin live stream broadcast interface" and you usually get something that looks like a tidy Western video call. The room is right, the host is right, and the screen is completely wrong: no danmaku, no gift tray, no PK bar, no half-transparent icon stack in the corners. Douyin's layout is dense in a way that most prompt writers never spell out, so the model fills the gaps with the interface it has seen most often.

This page fixes that. Below are 25+ Douyin live stream UI prompts grouped by the five broadcast types that dominate the platform, followed by a cheat sheet that translates each Douyin interface element into English you can paste into a prompt. If you want the cross-platform overview first, start with our AI livestream UI prompt collection, and if you also need the Chinese video-site look, the Bilibili live stream UI prompts page covers that separately.

Why Douyin UI Screenshots Are Hard to Fake

The first problem is density. A real Douyin screen stacks eight or nine layers: live video, danmaku lines crossing it, a viewer counter, a follow pill, a top-gifter avatar row, the gift tray, a comment input, a shopping cart badge and whatever event banner the platform is running that week. Tell a model "Douyin livestream" and it draws maybe two of those. You have to name them, and a good starting point is the way the classic livestream interface prompt names layout before it names anything else.

The second problem is direction and rhythm. Danmaku drift upward from the bottom-left with small round avatars attached, they are not a clean right-hand column like Twitch chat. Gift icons run horizontally along the bottom center. Leaderboards climb the right edge. Once you describe the direction of each layer, the whole frame snaps into place.

The third problem is that AI images hate Chinese characters. Ask for Chinese text on a UI and you get plausible-looking glyph soup, which is fine at thumbnail size and obvious at full size. The workaround is to keep labels to one or two characters, use a viewer count with a single Chinese unit like 1.2万, and let icons do the rest. If your screenshot needs a host in it, the lighting and framing advice in the prompt writing and editing guide applies here too.

Beauty and Fashion Live Stream Prompts

Beauty streams are the most common Douyin format and the easiest to over-light, so keep highlights soft and the UI pale. The clean live stream UI layout prompt and the artistic livestream overlay prompt are both useful references for a quieter frame.

Gaming PK Live Stream Prompts

A PK stream needs two things visible at once: two hosts and one score bar between them. This gaming stream overlay prompt and the simple livestream interface prompt show the split-tile idea without the PK props.

Live Shopping Stream Prompts

Shopping screens are judged on the price card and the cart badge, not the host. For the widget detail at full size, look at this product livestream host prompt and the shopping livestream icon prompt.

Talk Show and Music Live Stream Prompts

Talk and music streams live on mood and audio, so the interface should stay out of the way. Compare the dance livestream broadcast prompt and the interface plus livestream layout prompt for two different amounts of chrome.

Festival and Special Event Live Stream Prompts

Event streams are where Douyin leans hardest on red and gold, and where the gift animations get genuinely loud. This portrait livestream interface prompt is a plain frame to build the seasonal props on top of.

Douyin Live Stream UI Elements Cheat Sheet

Paste the phrasing from this list straight into your prompt. Two or three lines from here are usually enough to move an image from generic video call to recognizable Douyin screen.

How to Write Prompts for Douyin Screenshots

Start with the aspect ratio and resolution, because portrait is half the recognizability. Then describe the host and the room in one sentence, since the background does more for realism than any interface detail. Next list three or four interface elements by name from the cheat sheet, choosing the ones a real stream of that type would show: a PK bar belongs on a battle stream, a shopping cart badge belongs on a sale.

After that, describe the state of the screen rather than the design of it. Say that the danmaku are mid-scroll, that a gift animation is crossing the frame, that the PK bar is in the final ten seconds. A live screen looks slightly messy, and prompts that allow messiness come back looking used instead of rendered.

Finally, keep text minimal. Model-rendered Chinese characters are the give-away in almost every AI Douyin screenshot, so limit yourself to a viewer count and one or two single-character labels. If the result still looks off, the baseline frames in the interface prompt library and the livestream prompt library are useful references to compare against.

FAQ

Why does my AI Douyin screenshot look like a Western app?

Because you described the stream and not the interface. Adding a danmaku layer, a gift tray along the bottom and a top-left viewer count usually fixes it in one pass. Density is the difference, so more legitimate layers almost always read as more Douyin.

Can I generate readable Chinese text in the interface?

Rarely, and not consistently. Keep it to a viewer count with one Chinese unit plus a couple of single-character labels, and lean on icons everywhere else. Screenshots that were built around layout instead of lettering still work after a platform redesign.

Which prompt gives the most authentic PK battle screen?

The split-tile one above, with the score bar named explicitly. Add the final-ten-seconds state and a gift animation crossing the divider. If your model keeps merging the two hosts into one frame, describe each tile as a separate webcam feed with its own lighting.

How do I make a series of Douyin screenshots look consistent?

Freeze the host description, the room and the interface accent color, then change only the lighting and the stream state between prompts. Sets that drift in every variable look like unrelated generations, even when each individual image is fine.

Browse AI image prompts by model