seedance-game-ui-livestream

Gameplay capture with HUD and stream overlay · 游戏实机录屏与直播叠层

The screen itself is the shot: a fake gameplay capture, livestream or desktop recording. It holds up when the overlay layer is pinned to fixed positions and its numbers and banners change in step with the action.

This Skill carries one prompt structure distilled from 16 human-verified Seedance cases on goodcase.ai. It is a sibling of seedance-prompt-library (all templates in one Skill); install this one when you only want this kind of clip. Do not invent structure from general video-generation knowledge: follow the structure below and ground the draft in one anchor case from references/cases.md.

Use when

• GTA-style mission clips, streamer facecam plus game footage, and interactive desktop or UI recordings where the HUD has to read as a real interface.
• Chinese: For clips like GTA-style mission segments, a streamer's small window plus gameplay footage, or interactive desktop or interface screen recordings, the HUD needs to look like a real interface.

Workflow

1. Collect inputs. Ask only for what is missing: subject or product, setting, duration, aspect ratio, whether reference images / audio exist, and which language the final prompt should be in.
2. Pick one anchor case. Read references/cases.md, choose the case closest to the request, and name it in the reply. One anchor, never an average of several.
3. Fill the structure block by block. Every block below must be present in the final prompt; an empty block is where prompts go vague.
   1. Format header: duration, aspect ratio, how the take is cut, and a plain statement that this is game capture
   2. Character lock: Image1 for face and identity only, outfit written out in text
   3. Screen layer spec: where each HUD element, facecam, chat or subtitle sits, and what language it uses, fixed throughout
   4. Camera rig: third-person follow distance and FOV, or one locked camera for desktop recordings
   5. Timeline segments: each one carries the action, the HUD state change and any spoken line
   6. Audio: engine, footsteps, keyboard and mouse, ambience, voice language
   7. Strict rules tail: exact character counts, no extra cuts, HUD stays put, how the clip may and may not end
4. Apply the guidance and check the pitfalls (next two sections) before returning anything.
5. Return the finished prompt as one copy-pasteable block, the anchor case link, and a three-line checklist of what to verify in the generated video.

Guidance (from the cases)

• Pin every overlay to a named screen position before the timeline starts. GTA 6 Simulation opens with Fixed full-screen game HUD throughout and puts the streamer in a bottom-right square pink-blue neon facecam; the snow-station trailer assigns one element to each corner, stamina bars top-left, objective banner top-center, date top-right, minimap bottom-left, button prompts bottom-right.
• Treat the HUD as a scoreboard that changes with each beat. GTA 6 Simulation tracks ammo from 24/120 to 14/120 and wanted level from two stars to three; the diamond escape gives every segment its own HUD block, going from MISSION: STEAL VIP NECKLACE to TARGET ACQUIRED to ESCAPE SUCCESSFUL.
• Say outright that it is gameplay and write the camera like a game rig. The Rio chase asks for genuine gameplay, not a cinematic film; the diamond escape writes Clearly a GAME, not anime or cartoon and puts the camera 1.5m behind NAGI, slightly camera-right with FOV breathing between 30 and 60 degrees.
• Count the cast and make each extra person look different. GTA 6 Simulation asks for exactly two dark-red-jacket gang enemies and no extra armed characters; the five o'clock office case gives four coworkers different ages, heights and hair, and states that the boss is the only bald character.
• Split languages by layer and spell out how on-screen text appears. The diamond escape keeps All HUD text English, dialogue in Japanese; the desktop wallpaper case puts subtitles at the left middle of the frame and types them in at about 0.08 to 0.12 seconds per character, with a waveform under them.

Pitfalls

• The HUD slides around or changes layout between beats. Write HUD fixed in the same screen positions as a hard rule and only let the values change, never the layout.
• Long HUD sentences come out as garbled text. Keep banners to two to four capitalised words like the diamond escape does, and keep numbers in a simple pattern like 38/120.
• The streamer or hero shows up twice, once in the facecam and once in the game world, or in a reflection. GTA 6 Simulation states HANEUL appears only in the facecam; the office case removes any mirror that could create a second NAGI.
• The model edits it like a cinematic trailer, with cuts and a tidy ending. Write no cuts and no transitions; if you need a closing shot, declare one hard cut at an exact second, as in Exactly one hard cut at 27s, and rule out a black screen or end card.

Language

Reply in the user's language. The prompt itself can stay in English when that is what the user's Seedance workflow expects; ask if unclear. The Chinese version of this structure follows.

Chinese: Gameplay capture with HUD and stream overlay

The screen itself is the image, pretending to be a gameplay capture, livestream, or desktop screen recording. What makes it work is that the overlays are pinned to fixed positions, and the numbers and banners on them change beat by beat with the story.

Structure:

1. Format opening: duration, aspect ratio, how the camera cuts, then state plainly that this is a gameplay recording
2. Character lock: Image1 only handles the face and identity, write out the clothing fully in text
3. Screen overlay description: where each HUD element, streamer window, danmaku or subtitle goes, and in what language, fixed throughout
4. Camera setup: the distance and angle of a third-person follow, or for a desktop recording just write one fixed camera position
5. Timeline segments: for each segment, write clearly the action, the HUD state changes, and the dialogue spoken in that segment
6. Audio: engine, footsteps, keyboard and mouse, ambient sound, spoken language
7. Hard rules to close: lock the number of people, no extra cuts, HUD does not move, how it can and cannot end

Key points:

• Before the timeline starts, pin each overlay to a specific position. GTA 6 Simulation writes at the start Fixed full-screen game HUD throughout, the streamer goes in bottom-right square pink-blue neon facecam; in the snowy station one, one thing goes in each corner, stamina bar top left, mission banner top center, date top right, minimap bottom left, button prompts bottom right.
• Write the HUD like a scoreboard, and make it change once in every segment. GTA 6 Simulation wrote ammo from 24/120 to 14/120, wanted stars rising from two to three; in the diamond escape one, each segment has its own HUD block, from MISSION: STEAL VIP NECKLACE to TARGET ACQUIRED and then to ESCAPE SUCCESSFUL。
• State directly that this is game footage, and write the camera position according to the game camera. The Rio chase one requires it to look like a real gameplay recording, and wrote not a cinematic film; the diamond escape one wrote Clearly a GAME, not anime or cartoon, with the camera placed at 1.5m behind NAGI, slightly camera-right, and the angle breathing between 30 and 60 degrees.
• Lock the number of people, and every extra person must look different. GTA 6 Simulation requires exactly two dark-red-jacket gang enemies, no other gun-wielding characters may appear; the off-work-at-five one gives the four coworkers different ages, heights, and hairstyles, and specifically states that the boss is the only bald person in the whole film.
• Assign languages by layer, and also write clearly how the on-screen text appears. The diamond escape one specifies All HUD text English, with dialogue in Japanese; the dynamic wallpaper one puts the subtitles at the middle left of the frame, each character typed out over about 0.08 to 0.12 seconds, with a bouncing audio waveform below.

Common pitfalls:

• The HUD drifts between segments or changes layout. Write in the hard rules HUD fixed in the same screen positions, only let the values change, the layout never moves.
• Long sentences on the HUD come out as garbled text. Keep banners to two to four uppercase words, like in the diamond escape one, and use simple number formats like 38/120.
• The streamer or protagonist appears twice, once in the small window and once in the game world, or an extra one shows up in a mirror. GTA 6 Simulation states that HANEUL only appears in the bottom-right small window; the off-work-at-five one simply specifies that there is no mirror in the elevator.
• The model cuts it into a movie trailer, with cuts and a happy ending. Write clearly no cuts and no transitions; if you really want a closing shot, then like Exactly one hard cut at 27s pin the single cut to a specific second, and rule out black screens and end cards.

Copy-ready lead-in

For users who would rather paste into a chat model than run this Skill, hand them this line above the structure:

I want a video that looks like real gameplay capture. [The hero is a short-haired girl in a school uniform; I am sending you her photo.] [The mission: steal the last rice ball from a convenience store late at night and escape into the street.] [A streamer facecam sits in the bottom-right corner, with scrolling live chat on the left.] Using the prompt template below, rewrite it into one ready-to-use Seedance video prompt for me:

I want to make a video that looks like actual gameplay screen recording. [The protagonist is a short-haired girl in a school uniform; I will provide you with a photo of the person.] [The task is to steal the last rice ball from a convenience store late at night and then escape onto the street.] [There is a small streamer camera window in the lower right corner of the screen, and scrolling bullet comments on the left.] Please use the prompt template below to help me rewrite it into a Seedance video prompt that can be used directly:

Notes

• references/cases.md is generated from data/ in LearnPrompt/awesome-seedance and refreshed daily; do not hand-edit an installed copy.
• Prompts in the reference file belong to their creators (linked on every card). Transfer the structure; do not present a close copy as original work.
• For the full library across every model, install the goodcase Skill from LearnPrompt/goodcase-lite.