Band shells: Headline captions on the empty shell, not a generated crowd

Architecture plate of the shell. Headline is Bebas as a title row, kind subtitle, no transcript. Concert hours are overlay. Invented audiences are extras.

A band-shell clip that can run is a still of this shell — the arch, the stage lip, the park behind it — then a little motion, then the title and the hours as type you typed. It is not a generated lawn of faces. Photoreal audiences the model invented are extras you did not hire and a likeness problem you should not ship. Photograph the empty-enough shell you operate. Animate that. Print tonight’s line. Leave the crowd out of the generate.

The shell in the still has to be the shell in the park

Civic shells have a silhouette a local will recognise: the arch rib, the painted soffit, the steps, the trees that actually sit on that lawn. A paragraph that says “band shell at dusk, packed concert” will mint a cousin pavilion and a crowd you do not have releases for. Frame the real machine: same ribs, same lighting towers, same lake or street behind it. Dawn, empty, or staff-only is a legal still. A packed lawn of generated fans is not.

Ordinary image-to-video moves that photograph. It does not get to populate the foreground with people. Prompt the travel — slow clouds, light on the arch that is already in the still, flags if the flags are in the plate. Stop. “Crowd cheering” is a second job, and it is talent you did not hire. Keep lips and extras out of the generate.

/for/event-venues is a four-wall room and a floor plan. It is not this outdoor shell. Do not rewrite that page as a band shell, and do not import a venue crowd onto the lawn still. If the only photo includes a packed audience, crop to the architecture or shoot again. The shell is the subject. The empty bowl is the address.

Wan 2.5 I2V is the cheap silent hold on the locked plate

Wan Video 2.5 Image to Video is high-quality image-to-video. It requires an image. Audio is false. Durations: 5s and 10s. Ceiling 1080p. Display price 15 credits. Quality chips: 480p, 720p, 1080p. Motion levels: Medium, High. The still is the brief. The model moves it.

Fifteen credits is cheap enough to retry, not cheap enough to retry the wrong label. No still, no generate. A paragraph where the band shell should be is a wrong label. Every second of the 5s or 10s is waste — and a crowd the sampler will gladly invent if you ask.

The AI image-to-video tool is the door. Medium motion on an approved arch is often the whole job. High motion on a busy unapproved plate is ten seconds of confident drift and extra limbs. 1080p will faithfully resolve whatever you uploaded, including a generated extra you should not have prompted in.

Silent is the spec. Prompting a brass fanfare here does not buy a bed. Score later. Hours are overlay. Headline is not Wan glyphs. Do not “fill the lawn” in the prompt to make the still look open. Open is a photograph of a real concert, with releases in the folder. Generation is not a crowd-size slider.

Headline is Bebas as a title row, not a transcript

Headline is Bebas Neue, tall caps, position top, kind=subtitle in the registry but the job is a hook line. Do not feed it a transcript. The family reads as a title card or a news chyron. It is not karaoke. It is not a word-pop on spoken audio.

The base is bebas_top: fontId bebas, white fill, black outline width 1, position top, kind subtitle. The face itself is Bebas Neue — a display caption/overlay face, not a generate. Tall condensed caps. No lowercase. Five variants share the face: top, bottom, yellow fill, red outline, center. They change colour and seat. They do not grow a transcript.

kind=subtitle means the line arrives together. It does not mean “dump the concert PA into this family.” Auto-captioning a mute shell hoping “TONIGHT” appears is a spend on silence. Overlay the title you wrote: the series name, the act, the date. Asking Wan to “put a Bebas headline in the shot” is painted glyphs you cannot edit.

Concert hours are overlay; invented audiences are extras

Add a text overlay to a video stamps one authored line on the whole file — top, center, or bottom. “Gates 6. Music 7. Rain or shine.” That string changes. The shell does not. Burn hours after picture lock and after the crop. A 16:9 overlay that then gets mashed into Stories sits in the chrome.

If the line has to change on a clock you can name — act, then time, then a rain line — time the overlays. Still authored. Still not a transcript. Recut after this pass and every timestamp sits on the wrong downbeat.

A cinematic talking row will happily build a packed lawn from a sentence. That packed lawn will include faces the model invented. Do not do that. A silent shell on Wan is the architecture job. A real photographer on concert night is the people job. There is no third job called “generate the crowd so the park looks busy.”

If a season clip needs faces, those faces are patrons you have permission to show, on footage you shot. Generation stays on the arch, the empty bowl, the hours. If a local can name the shell, you locked the right still. If a parent might ask who that person on the blanket is, you shipped an extra. Print the time. Leave the lawn as you photographed it.

FAQ

Can we add a generated crowd so the shell looks full?

No. Photoreal extras are a likeness problem. Photograph a real concert with releases, or keep the still empty-enough and overlay hours. Invented audiences are not scale.

Does Wan 2.5 I2V include a concert soundtrack?

No. Audio is false. You get a silent 5s or 10s file at up to 1080p for 15 credits. Score after lock. Hours are overlay, not Wan glyphs.

Why is Headline kind subtitle if the job is a title?

kind=subtitle is the registry class: the whole line appears as a block. There is no word-timed effect. The job is still a hook you authored in Bebas Neue. Do not read subtitle as “dump the PA transcript here.”

Is this the same as the event-venues page?

No. /for/event-venues is indoor four-wall bookings. This page is the civic shell: empty architecture, Headline title row, hours as overlay, no generated crowd.