A faceless narrated channel: the tools, the workflow, the next episode

What goes wrong, and the fix

Before you start

• A niche in one line. Ours: short stories about cities after dark, for people who cannot sleep.
• A look in one sentence. Ours: anamorphic night photography, amber sodium streetlight and warm practical light on wet streets, light haze, fine film grain.
• A character with a saved catalog voice for the narrator. Ours is Isla Marr, the narrator from The same voice in every video: a generated portrait with the ElevenLabs v3 voice Sarah.
• Pro or above for the video models. Speech runs on every plan.

Anamorphic night photography, amber sodium streetlight and warm practical light on wet streets, light haze, fine 35mm film grain, deep blacks, cinematic contrast.

Write the script to the clip length

Each episode here is two five second clips, so each script is two sentences, one per clip. Write it before you make anything else; the narration and the shot list both come from it.

Every city has a tram that runs after the timetable ends. Nobody gets on. That is not the point.

At three in the morning the city is not asleep, it is breathing through its rooftops. That hum is a thousand rooms staying warm.

Read it with the saved voice

Open the audio composer on the Voiceover task, press Speak as, pick the narrator, paste the episode's script and press Generate audio. One take per episode; the button shows the price for exactly those characters. Ours came back at 6.2 seconds for episode 1 and 8.5 seconds for episode 2, both under the ten second picture.

Make the b-roll in one look

Open the video composer, pick Seedance 2.5 in the model picker, press the audio switch until it reads Audio Off, keep five seconds, 16:9 and 720p, and write each shot as the look sentence followed by one continuous shot. The button shows the live price for those settings. Two shots per episode. All four shots are Seedance 2.5 with these settings, and each one is its first run: the tram came back empty, with no driver in the cab, the first time. Our prompts end with no signs and no lettering because this channel's streets are empty by design; if your shot needs a sign or a label, say exactly what it reads, or upload the logo image.

Anamorphic night photography, amber sodium streetlight and warm practical light on wet streets, light haze, fine 35mm film grain, deep blacks, cinematic contrast. One continuous shot: an empty night tram glides slowly along a wet street toward the camera, its windows glowing warm amber and showing empty seats, no passengers and no driver. Bare wet stone facades on both sides lit only by tall sodium streetlamps. No signs, no lettering, no posters, no people anywhere in the frame. Slow low-angle tracking shot.

Anamorphic night photography, amber sodium streetlight and warm practical light on wet streets, light haze, fine 35mm film grain, deep blacks, cinematic contrast. One continuous shot: inside an empty night tram, rows of worn seats under warm amber light, raindrops on the window, the city sliding past outside in long streaks of orange sodium streetlight, slow dolly forward down the aisle. No text, no logo, no readable signs.

Anamorphic night photography, amber sodium streetlight and warm practical light on wet streets, light haze, fine 35mm film grain, deep blacks, cinematic contrast. One continuous shot: a city rooftop at three in the morning, rows of humming air-conditioning units under orange sodium light and one cold work lamp, thin steam rising and drifting, the blurred skyline behind, slow lateral tracking shot. No text, no logo, no readable signs.

Anamorphic night photography, amber sodium streetlight and warm practical light on wet streets, light haze, fine 35mm film grain, deep blacks, cinematic contrast. One continuous shot: an empty late-night diner seen through rain-streaked glass, warm pendant lights glowing over the counter and one flickering fluorescent tube above the grill, a slow push-in through the reflections, amber streetlight from the street behind the camera. No text, no letters, no logo, no readable signs.

Cut it in Studio

Open Timelines, press New timeline, put it in the channel's project and open it. In Studio's left panel switch from Project media to Library, search for the shot and press the + on its card: shot one lands at 0. Press Jump to end on the transport and press + on shot two, so it starts where shot one ends. Drag the scrubber back to where the first word should land (0.8 seconds in episode 1) and press + on the narration take. Each + lands a clip on a fresh track; ours sit on one video track, and the cut plays the same either way. Then the score and the ambience, below.

The score. Make it in the audio composer's Music task, asking for a quiet stretch as long as the narration and a lift after it. Play it in the Source preview, press Mark In where the quiet stretch starts and Mark Out ten seconds later, move the playhead to 0 and press +. These timelines were made today, so the export dips the score under the voice and evens the loudness, about -14 LUFS. The Audio inspector sets any sound's level and fades. Then add the ambience: a tram rolling past on wet rails under shot one and light rain under shot two, each a short Sound effect.

Melancholic late-night cinematic score for a short story about an empty night tram, 70 bpm, soft felt piano, warm sub pad, distant brushed percussion, instrumental, no vocals. Very quiet and sparse for the first eight seconds, then a gentle swell of strings, then quiet again, and that shape repeats.

Very quiet, steady nocturnal ambient pad for a city at three in the morning: a soft warm synth drone, a faint slow pulse, a distant hum, gentle and continuous from start to finish, low level, instrumental, no vocals.

We measured both exports: no gap of silence longer than half a second (0.2 s at most), and the voice well above the score and the ambience while she speaks: in nine of ten 50 ms windows of her speech it is at least 6.8 dB above everything else (8.6 dB in episode 2).

Hand the next episode to the agent

The NOLGIA Agent holds the channel's context in a chat: the niche, the look sentence, the narrator's name. We started a new chat with the brief below, ended it with a budget line for the turn (you can read it in the screenshot), and asked for episode three's script and shot list only. It answered in one turn with a title, two sentences and two shot prompts in the look sentence, and generated nothing, so the turn cost the per turn rate of its brain and nothing else. It also remembered an earlier episode 3 draft from another chat on the account, in the old look, and said this one replaces it. From there you can ask it to make the clips and the take and file them in the project, or run the shots yourself.

You are the writer for After Hours Atlas, a faceless channel of short stories about cities after dark for people who cannot sleep. The narrator is the character Isla Marr. The look sentence for every shot is: Anamorphic night photography, amber sodium streetlight and warm practical light on wet streets, light haze, fine 35mm film grain, deep blacks, cinematic contrast. Write episode 3 only: a title, a two-sentence script (one sentence per five second shot), and the two shot prompts, each starting with the look sentence and ending with no text, no logo, no readable signs. Do not generate anything yet.

Which tool for which job

Where you still regenerate

• A look sentence is not a lock. Two shots in the same look still differ in grain, palette and framing, and an object can come back as something else: our rooftop units came back as skylight hoods.
• Levels. The export dips the score under the voice and evens the loudness. If the score still buries a line, lower its Clip volume in the Audio inspector, or move its Mark In to a quieter passage.
• Lettering. A street scene can grow signs you did not ask for, and generated lettering can come out garbled. Say exactly what a sign or label reads, or upload the logo image; if lettering comes out garbled, regenerate that shot. Check every clip at full size before the cut.
• Posting is yours. NOLGIA does not schedule or publish to YouTube or TikTok. Download the MP4 and post it.
• No income promise. This guide is about making episodes, not about what a channel earns.

What it costs

One ten second episode here is two clips, one narration take, one score, three short sound cues and an unbilled export. The preset and the explainer show their own estimates. Live rates:

A chat turn with the NOLGIA Agent bills at the rate of its brain, shown on the Agent page.

Try it yourself

A shot you make from our prompt will differ from ours; the look sentence holds the palette and the mood, not the frame. The tram came back right on our first run; yours may not.

Checklist for every episode

1. Script: one sentence per shot, written first.
2. Narration: Speak as the narrator, one take, length under the picture.
3. B-roll: look sentence first, one continuous shot, audio off.
4. Inspect every clip at full size, and regenerate any shot whose lettering comes out garbled.
5. Score: asked for with a quiet stretch as long as the narration; ambience as short cues.
6. Studio: Library tab, shot one with +, Jump to end, shot two with +, the take at the playhead, the score window at 0, the cues, Export video.
7. Save the script and shot list in the project for the next one.