NOLGIA

View saved version →

---
name: animate-a-logo
description: Use this when someone wants their logo or wordmark turned into a short animated ident, sting, bumper, reveal or end card. Trigger on logo animation, animate my logo, logo reveal, brand sting, ident, intro. Do not use it to design a new logo from nothing, and do not use it for a full advertisement.
---

# Animate a logo

Take a still mark and hand back a short ident the person can put on the front or the back of anything.

## When to trigger

Trigger on "animate my logo", "logo reveal", "I need an intro sting", "make an end card", "a bumper for my videos".

Do not trigger when there is no mark yet and the person actually wants one designed. In that case offer the logo route in run-a-preset first, then come back here.

## Ask for this first

Ask once, bundled:

1. A link to the mark, or confirmation it is already in their NOLGIA library. A version on a plain background works best.
2. How long, usually three to six seconds.
3. The feeling: clean and corporate, fast and sharp, warm and handmade, heavy and cinematic.
4. The shape: wide, square or vertical.
5. Whether a tagline appears under it, and the exact wording if so.

## Steps

1. If the mark is already theirs, call `nolgia_list_assets` and `nolgia_get_asset` to get a fresh link. If they only have it on a busy background, offer to clean it first with `nolgia_image_to_image`.
2. Call `nolgia_list_presets` if it has not been called yet in this conversation, and check the logo
   reveal routes. If one matches the feeling they described, follow its description.
3. Call `nolgia_list_models` and pick a video model that accepts a starting frame at the length and shape you need.
4. Call `nolgia_image_to_video` with `image_url` set to the mark. Write the movement yourself: how the mark arrives, what the light does, how it settles, and how long it holds at the end. Say in the prompt that the mark must match the reference exactly, with no redesign and no added or altered lettering. Set `aspect_ratio` and `duration_seconds`.
5. Poll `nolgia_list_jobs` until the job finishes, then call `nolgia_get_asset`.
6. Check the result frame by frame before delivering. Reject anything where the mark warps, the lettering changes, or extra text appears. Re-run it.

## What to return

The ident, its length and shape, and a note on where it sits best, front or back. Offer a second version at a different length, since most people want both a long and a short one.

Never name a model and never discuss what anything cost.

## When to stop or decline

- No mark and no willingness to make one: stop.
- The mark belongs to a company the person does not represent: decline.
- The lettering comes back wrong twice in a row: say so, and suggest holding the mark still and animating the background instead.

View saved version →

---
name: every-angle-from-one-take
description: Use this when someone has one steady video take, filmed from a single fixed camera, and wants it as other camera angles, such as a close up, a wide, a top down view or a slow push in, all in sync over the original sound, or wants a highlight edit that cuts between angles. Trigger on multicam, more angles, coverage, another angle of this, make it look like it was shot with several cameras. Do not use it to change what happens in a clip or to change its look.
---

# Every angle from one take

Multicam turns one steady take into the camera angles the person picks, each one a clip as long as the take and in sync over its own sound, plus a highlight edit that cuts between them.

## When to trigger

Trigger on "give me more angles of this", "make this look like a multicam shoot", "I only had one camera", "can I get a close up of this take", "cut a highlight between angles".

Do not trigger when the person wants the action, the subject or the setting changed (use remake-a-clip), only the look changed (use grade-my-footage), or a new shot made from a brief (use make-a-video).

## Ask for this first

Ask once, bundled:

1. Which take: a clip already in their NOLGIA library, or one they will add there first. It must be one fixed camera and at most 2 minutes long.
2. What it is for, if they want a ready-made set: a music video, scene coverage, an ad, a fashion film, a performance or dance, or an interview.
3. Any angle they already know they want, in their own words if it is not a standard one.

## Steps

1. Find the take with `nolgia_list_assets` filtered to video. Where the host shows this server's cards, let the person point at it rather than reciting ids.
2. Call `nolgia_multicam_plan` with the take, plus the pack or angles they asked for. It is free and renders nothing. Where the host shows this server's cards the plan appears as a picker with the price, so do not read the angle list back; say in one sentence what the take is and what the plan suggests.
3. If they change their choice, call `nolgia_multicam_plan` again with the new choice so the price shown is the price they will pay.
4. Only when the person has chosen the angles and agreed to the price, call `nolgia_multicam_start` with exactly the same take, pack, angles, custom angles, quality and notes as the plan they agreed to. Never start a job they have not agreed to.
5. The job keeps running after the chat moves on. Check it with `nolgia_multicam_run` about every half minute while it is running, not in a tight loop, and tell the person roughly how long it takes.
6. When angles are finished, offer everything in one file with `nolgia_multicam_download`. Give the link exactly as returned.
7. If they ask to stop, call `nolgia_multicam_cancel` after confirming, and tell them what the result says about finished angles and credits.

## What to return

The finished angles and the highlight edit, each named by angle, and the one download link when they want the files together. Offer one next step: another angle, or the highlight edit cut to a different length in the NOLGIA editor.

Never name a model, and never promise a refund or a price the tool result does not show.

## When to stop or decline

- No take, or a take that moves the camera or cuts between shots: explain that Multicam needs one steady shot from a single camera and ask for one.
- A take longer than 2 minutes: ask them to trim it first.
- The plan hides an angle because there is no clear person in the take: say so plainly and offer the angles the plan does show.
- The balance does not cover the price: say so and stop; nothing starts.

View saved version →

---
name: grade-my-footage
description: Use this when someone wants the look of a clip they already have changed rather than new footage made, such as a color grade, a film look, a warmer or cooler feel, a day for night pass, or matching two clips to each other. Trigger on grade, color grade, LUT, film look, match the look, fix the color. Do not use it to generate a new scene.
---

# Grade my footage

Change how an existing clip looks without changing what happens in it.

## When to trigger

Trigger on "grade this", "give it a film look", "make it warmer", "these two clips do not match", "can you fix the color", "apply a LUT".

Do not trigger when the person wants a new shot, a different action or a different subject. That is a new render, so use make-a-video.

## Ask for this first

Ask once, bundled:

1. A link to the clip, or confirmation it is already in their NOLGIA library.
2. The look they are after, in their own words. A reference film, a season, a mood or a named look all work.
3. Whether anything must stay exactly as it is, such as skin tone, a brand color or a product finish.
4. Whether this has to match another clip, and where that one is.

## Steps

1. Call `nolgia_list_color_presets` to get the names of the built-in looks. Use those names as vocabulary when you talk to the person, so they can point at one instead of describing a feeling.
2. If the clip is already theirs, call `nolgia_list_assets` filtered to video and `nolgia_get_asset` for a fresh link.
3. Call `nolgia_list_presets` if it has not been called yet in this conversation, and read the color
   grade route, which is the guided version of this job.
4. Call `nolgia_list_models` and pick a video model whose `references` block accepts video references, then check how many it takes.
5. Call `nolgia_text_to_video` with the clip passed as a video reference, either `video_asset_ids` for something in their library or `video_urls` for a link. Write the grade in the prompt: the contrast, where the shadows sit, the color of the highlights, the saturation, and the grain if any. State plainly that the action, the framing and the cut must not change, only the look. Name anything that must be protected.
6. Poll `nolgia_list_jobs`, then `nolgia_get_asset`.
7. Compare the result against the original before delivering. If the action changed, the framing moved or a face changed, that is a failed grade. Say so and re-run with a tighter instruction.

## What to return

The graded clip alongside a plain description of the look you applied, so they can ask for more or less of it. If they are matching two clips, deliver both and say which one you matched to.

Tell them the exact built-in looks can also be dialled in by hand in the NOLGIA editor if they want frame level control. Never name a model and never discuss what anything cost.

## When to stop or decline

- No clip: stop and ask for one.
- The clip is long enough that a full re-render is the wrong tool: say so, and point them at the editor for a grade applied over the whole timeline.
- The request is to make real footage of a real person look like something that did not happen: decline.

View saved version →

---
name: make-a-video
description: Use this when someone wants a finished video made from a written brief, such as a social ad, a brand spot, a product film, a scene, an explainer or a trailer, and also when they hand over a still image and want it to move. Do not use it to change the look of footage that already exists, and do not use it for narration or music on its own.
---

# Make a video

Turn a spoken brief into a finished clip the person can download and post.

## When to trigger

Trigger when someone describes a video they want and expects a file back. Typical openings: "make me a 15 second ad for", "I need a spot for my brand", "turn this photo into a clip", "can you film this scene".

Do not trigger when the person only wants an existing clip regraded (use grade-my-footage), when they only want narration or music (use voiceover-for-a-script), or when they are asking what NOLGIA can make (use run-a-preset).

## Ask for this first

Ask once, in a single friendly message, and accept one word answers:

1. What it is for, and the brand, product or subject.
2. Where it runs, so the shape is right: vertical for phones, wide for a site or a screen, square for a feed.
3. How long, in seconds.
4. The feeling, in three words.
5. Anything of theirs it must include: a link to a photo, a product shot, a logo, or an asset already in their NOLGIA library.

If they say they do not know, offer two or three concrete directions and take it from there. Never make them write the prompt. You write the prompt from their answers.

If they give a complete brief up front, follow it to the letter and only fill the gaps.

## Steps

1. Call `nolgia_list_presets` ONCE, and only if you have not already called it in this conversation: a
   repeat returns the same catalog and draws a duplicate gallery. If one of the guided routes clearly
   matches what they asked for, say so and hand over to the run-a-preset skill instead of improvising.
2. Call `nolgia_list_models` and pick a video model whose durations and aspect ratios cover the brief, and whose reference inputs cover whatever material they supplied. Check the model's `references` block before you pass any reference field, because the server rejects fields the model does not accept.
3. If they gave a starting image, call `nolgia_image_to_video` with `image_url` set to that image. Otherwise call `nolgia_text_to_video`. Write a full prompt yourself: the setting, the action, the camera, the light and the mood. Set `aspect_ratio` and `duration_seconds` from their answers. For anything longer than one continuous take, pass `shots`, each with its own `prompt` and `duration_seconds`.
4. Video runs as a background job. Poll `nolgia_list_jobs` until the job is finished or failed. Tell the person it is rendering and roughly how long it takes. Do not poll in a tight loop.
5. When it finishes, call `nolgia_get_asset` for the link.
6. If this is part of a campaign with more pieces coming, call `nolgia_create_project` once and then `nolgia_add_project_assets` so everything stays together.

## What to return

The finished clip, a one line description of what you made, and the shape and length. Offer one next step: a cutdown, a different aspect ratio, or a voiceover over the top.

Never name a model, never quote a preset's internal guidance text back to the person, and never discuss what anything cost.

## When to stop or decline

- No brief and no answers: ask, do not guess.
- They want a real, named person depicted: decline and offer a character of their own instead.
- The job comes back failed: say so plainly, say what you will change, and offer to re-run once. Do not re-run silently more than once.

View saved version →

---
name: make-it-with-encore
description: Use this when someone wants a finished, styled edit made from their own files and words in one of the Encore styles, such as a product drop, a travel film, a lyric or music video edit, a fit check lookbook, a car reveal, an anime hero opening, a talking clip cut like a trailer, captions for a talking clip, or an edit they love remade with them in it. Trigger on Encore, make it in this style, paste the edit, remake this edit with me, turn my footage into a travel film, launch edit for my product. Do not use it for a single new shot from a brief, for changing the look of one clip, or for other camera angles of one take.
---

# Make it with Encore

Encore makes a finished edit in one of its styles from the person's own photos, clips, song and words, on NOLGIA's servers. It keeps going after the chat moves on, and the result lands in their library and in Studio.

## When to trigger

Trigger on "make an Encore", "make me a product drop", "turn my trip footage into a travel film", "a lyric video from my song", "remake this edit with me in it", "caption my talking clip", or any Encore style by name.

Do not trigger for one new shot from a brief (use make-a-video), a new look on one clip (use grade-my-footage), a remake of one clip with a new subject or place (use remake-a-clip), or other camera angles of one take (use every-angle-from-one-take).

## Steps

1. Call `nolgia_encore_catalog` to show the styles. It takes no arguments. Where the host shows this server's cards, the styles appear on a card with posters and previews; say in one sentence what Encore does and let the person pick.
2. Call `nolgia_encore_style` with the chosen `preset` (and a `mode`, if the person picked one) for its form. Each mode says what it needs: photos, clips, a song, a saved character, words.
3. Gather what the form needs.
   - Files must be in the person's NOLGIA library. Use `nolgia_list_assets` to find them. Save files attached in this chat with `nolgia_add_chat_files`.
   - For the flagship's edit given as a direct link, use `nolgia_encore_link_clip`, only after the person says they made the clip or hold its rights.
   - Never invent the person's words. Ask for them, within the limits the form shows.
4. Price the form with `nolgia_encore_quote`. It prices a new run of one style and is free. Pass on every problem it names in its own words, fix them, and price again.
5. Ask the person to confirm the consent statement the form shows. It says that everyone in their files agreed to appear and that they hold the rights to the files. Also ask them to agree to the price. A style cut from their own footage often costs nothing; a generated one says its price and the retake reserve.
6. Only then call `nolgia_encore_start` with the same form, `consent` true and `agreed_credits` set to the quoted price. Nothing starts if the price moved, the balance is short or the form has a problem.
7. Check the run with `nolgia_encore_status` about every half minute while it runs, not in a tight loop. Where the host shows cards, the card refreshes itself.
8. When it finishes, the edit plays on the card with its compare file. Offer `nolgia_encore_download` for every file in one ZIP, and give the link exactly as returned.
9. If they ask to stop, confirm first, then call `nolgia_encore_cancel` and say what the result says about charges.

## Making shots of a finished run again

When a finished run's result lists a residual with a `redo` list (for example a sign or a logo in one shot of an Encore Travel film that reads wrong), those shots can be made again without making the whole run again.

1. Price it with `nolgia_encore_redo_quote`: the finished run's `run_id`, the shot ids from the residual's `redo` list as `items`, and, if the person has them, up to 3 library images of the logo or text as `vibe_marks`. It is free. Pass on any problem in its own words.
2. Only once the person has agreed to the price, call `nolgia_encore_redo_start` with the same `run_id`, `items` and `vibe_marks`, `consent` true and `agreed_credits` set to the quoted price. Every other shot of the finished run is kept as it is.

## For the flagship, "Paste the edit"

The flagship (preset `encore`) remakes an edit the person made or holds the rights to, with them in it.

1. Scan the edit first with `nolgia_encore_scan_edit`: the clip's library id as `reference`, and `consent` true only once the person has confirmed the consent statement. The scan is free and makes no edit. Check it with `nolgia_encore_status` until it finishes; the finished scan lists the edit's word slots.
2. Then price the remake with `nolgia_encore_quote`: preset `encore`, mode `remix` (or `exact` to keep the edit's own moves shot for shot), the finished scan's run id as `inputs.scan_id`, one subject (`subject_photos`, `character_id`, `product_photos` or `footage`), and the person's own words keyed by the scan's word ids. Start it with `nolgia_encore_start` as in the steps above.

## When to stop or decline

- A link to TikTok, Instagram, YouTube, X, Facebook, Vimeo or a similar platform: do not fetch it. Ask for a clip they made, attached here or uploaded to nolgia.ai.
- Files that show people who did not agree to appear, or a song or edit they do not hold the rights to: stop and say why.
- A real, named public figure as the subject: decline and offer their own photos or a saved character instead.
- The balance does not cover the run: say so and stop. Nothing starts.

Never name a model, and never promise a price the tool result does not show.

View saved version →

---
name: product-shots
description: Use this when someone has photographs of a real product and wants a consistent set of images out of them, such as product photos, catalogue shots, packshots, marketplace listing images, an Amazon main image, A+ content, an infographic, a white background set, a hero banner, lifestyle scenes or a seasonal refresh. Do not use it when there is no source photograph of the product, and do not use it for video.
---

# Product shots

Take the photographs a person already has and return a set that looks like it came from one shoot.

## When to trigger

Trigger on "product photos", "product shots", "packshots", "listing photos", "listing images", "Amazon listing", "main image", "A+ content", "infographic", "hero banner", "carousel set", "catalogue set", "on white", "lifestyle shots of my product", "make these look professional", or on someone sharing product photographs and asking for more.

Do not trigger for a video of the product (use make-a-video or the product demo route in run-a-preset), and do not trigger when the product does not exist yet.

## Ask for this first

Ask once, bundled:

1. A link to the photographs, or confirmation that they are already in their NOLGIA library. Up to four references can ride along on a single image.
2. The product name and anything that must read correctly, such as label text or a finish.
3. The set they want: plain white, a surface and a prop, a lifestyle scene, or all three.
4. The shape: square for a listing, vertical for a feed, wide for a banner.
5. How many.

The server needs a link it can open. If the person attaches a file to the chat rather than giving a link, tell them plainly that you need either a link or an image already saved in their NOLGIA library, and offer to work from the library.

## Steps

Presets first. Call `nolgia_list_presets` once (reuse the result if it is already in the conversation). If the catalog carries `product-photos` (hero shot on white, lifestyle, closeup with hands, hero banner, carousel directions) or `marketplace-listing-images` (main image, infographic, multi-angle, detail shot, lifestyle, what's in the box), open the one that fits with `nolgia_get_preset`, pick the direction that matches their words, bind the photographs to its reference field and run it with `nolgia_run_preset`. Only when neither is in the catalog do the direct steps below apply.

1. If the images are already theirs, call `nolgia_list_assets` filtered to images, then `nolgia_get_asset` on the ones they mean to get fresh links.
2. Call `nolgia_list_models` and pick an image model that accepts the number of references you plan to send. Several models take only one reference.
3. Call `nolgia_image_to_image` once per shot in the set. Pass the product photograph as `image_url`, and additional references as `image_urls` only on a model that accepts them. Set `aspect_ratio` from their answer. In the prompt, name the product exactly as it appears in the reference and require that it match the reference exactly, with no redesign, restyle, recolor or reinterpretation. Describe only the set, the light and the angle.
4. Look at every result before you deliver it. If the label, the shape, the color or the finish drifted from the reference, re-run that shot alone. Do not deliver a drifted product.
5. Call `nolgia_create_project` for the set, then `nolgia_add_project_assets`, then `nolgia_update_asset_tags` with something the person will recognise later, such as the product name.

## What to return

The set, grouped, with one line saying what each shot is for. Say which reference each came from. Offer one more angle or one more background.

Never name a model and never discuss what anything cost.

## When to stop or decline

- No usable source photograph: stop and ask for one.
- The photographs show a person's face and the person is not the account holder's own model: say you will keep the product in frame and crop or replace the person rather than reproducing them.
- The product carries another company's branding and the request is to pass it off as the person's own: decline.

View saved version →

---
name: remake-a-clip
description: Use this when someone has a video clip and wants it remade rather than a new one made from scratch. Three ways are covered. Keep the clip's movement and have their own photos, characters or product perform it, swap or remove one person, animal, product, prop or outfit and keep the rest of the shot exactly as filmed, or keep the subject and its movement and move the whole scene to a new place with new light. Trigger on remake, redo this clip, swap, replace the person, remove the boom mic, take the crew member out, get rid of the cup, put my product in this, same moves, new setting, new world. Do not use it to change only the colors of a clip.
---

# Remake a clip

Miroge remakes a clip the person already has. It keeps what they want from the original and changes the rest.

## When to trigger

Trigger on "remake this with my product", "same dance, but it's me", "swap the mug for our bottle", "remove the boom mic from this take", "take the guy in the vest out of the shot", "put this scene on a beach at sunset", "keep the rider, change the city".

Do not trigger when the person only wants the colors or the grade changed (use grade-my-footage), wants other camera angles of the same take (use every-angle-from-one-take), or has no clip at all (use make-a-video).

## The three ways

- **Motion transfer**: the clip's moves, timing and camera, performed by the person's own photos, saved characters or product.
- **Swap**: replace one person, animal, product or outfit and keep everything else as filmed. Swap can also **remove** one thing (a boom mic, a crew member, a stray cup): Miroge finds it in the clip, redraws only that spot, and gives back every other frame exactly as filmed, faces and sound included.
- **New world**: keep the subject and its moves, and redraw the place and the light around it.

## Ask for this first

Ask once, bundled:

1. Which clip: one already in their NOLGIA library.
2. Which of the three ways, in their words if they are not sure. You can map it.
3. For motion transfer or a swap, who or what goes in: a photo in their library, a saved character or a saved product. A swap can also be described in words. For a removal, only what should go, in a few plain words, the way it looks in the shot.
4. For a new world, the new place, its material, color and light, and anything that must stay exactly as it is.

## Steps

1. Find the clip with `nolgia_list_assets` filtered to video, and any photos they name the same way. Where the host shows this server's cards, let the person point at them.
2. If they want to compare prices or models first, call `nolgia_miroge_options` with the clip, the mode and the same photos, characters or product. It is free.
3. Call `nolgia_miroge` with the clip as `source_asset_id`, the `mode` (motion, swap or world), the cast in `element_asset_ids`, `character_ids` or `product_id`, and the person's own words in `prompt`. Miroge writes the rest of the instruction itself, so do not pad the prompt. For a new world, put what must not change in `keep`. To take something out, send `mode` swap with `remove` naming it in a few plain words ("the boom mic", "the man in the yellow vest") and no cast: that is what the site's Remove sends, and everything else comes back exactly as filmed. Leave length and shape alone unless the person asks, so the remake matches the clip.
4. The render runs in the background. Check it with `nolgia_get_miroge`, passing the same clip and mode so the card shows the original beside the remake. Do not poll in a tight loop.
5. Only if the person asks for another version of the identical remake, send it again with `another_take` set. Otherwise a repeat within five minutes returns the render already running.
6. If they ask to stop it, call `nolgia_cancel_job` after confirming, and tell them what the result says about credits.

## What to return

The remake beside the original, the way it was remade in one line, and what was kept. For a removal, say what was taken out and that the rest of the shot, faces and sound are the original's. Offer one next step: the same clip in one of the other two ways, or a different place for the new world.

Never name a model, and never promise a refund or a price the tool result does not show.

## When to stop or decline

- No clip: stop and ask for one.
- A new world with no description of the place: ask for it before anything is spent.
- The swap or the cast would put a real, named person the user has no rights to into the clip: decline and offer a saved character of their own instead.
- The render fails: say so plainly with the reason the result gives, and offer one re-run. A removal that could not find the thing says what it looked for and refunds the credits: ask the person to name it more plainly (its color and what it is) before running it again.

View saved version →

---
name: run-a-preset
description: Use this when someone asks what NOLGIA can make, names one of the guided routes such as a commercial, a social media ad, a product demo, product photos, listing images, an unboxing, a music video, a logo reveal, a poster, a cover, animated titles or a color grade, or wants help choosing where to start. Do not use it when the person has already given a complete brief for a single piece and just wants it made.
---

# Run a preset

Presets are NOLGIA's guided routes. Each one knows what it needs, so the person answers a few short questions instead of writing a brief.

## When to trigger

Trigger on "what can you make", "show me the options", "do you have something for", or on any named route: commercial, social media ad, creator review ad, app ad, product demo, product commercial, outfit try-on, unboxing reaction, narrated channel video, fashion film, music video, film scene, animated cartoon, motion graphics, animated titles, logo reveal, logo, poster, billboard, cover art, YouTube thumbnail, interior design, concept art, color grade footage, product photos, listing images, A+ content, infographic.

Do not trigger when the person has handed you a finished brief for one piece. Make that instead.

## Steps

1. Call `nolgia_list_presets` ONCE. It takes no arguments and returns the whole catalog, so if it has
   already been called in this conversation, reuse that result: calling again draws a second, identical
   gallery on top of the first.
2. Narrow to at most three that fit what the person said, using each preset's own name, description and best-for line. Prefer the ones whose output type matches what they asked for, a still or a moving picture.
3. The catalog is ALREADY ON THEIR SCREEN as a gallery of cards they can scroll, filter and open. Do not
   list it back to them. Name at most three that fit, one line each on what it makes and who it suits,
   and ask which one. If they already named one, skip straight to it.
4. If the preset card shows a FORM, the person is already looking at a field, chips or a library picker
   for every question. Point at it and let them fill it in rather than asking the same things in chat.
   Only when the card says a step needs answering in the conversation do you ask it yourself.
   Otherwise, ask the preset's questions in one bundled message. Every preset needs the subject, where it runs, and the feeling. Add the ones the chosen route obviously needs: a tagline for a commercial, a hook line for a social ad, the product and its photos for a demo, the mark for a logo reveal, the track for a music video, the source clip for a color grade.
5. Write all the prompting yourself from their answers and the preset's description. Their words are inputs, never the prompt.
6. Generate. Stills go through `nolgia_text_to_image`, or `nolgia_image_to_image` when they supplied a reference. Moving pictures go through `nolgia_text_to_video`, or `nolgia_image_to_video` when there is a starting frame. Check `nolgia_list_models` first so the model you choose accepts the length, the shape and the references you plan to send.
7. Video is a background job: poll `nolgia_list_jobs`, then `nolgia_get_asset` for the link. Stills come straight back.
8. If the person supplied a photo, a product, a garment or a mark, repeat it word for word in the prompt and say it must match the reference exactly. Compare the result against what they gave you before you deliver, and re-run anything that drifted.

## What to return

The finished file, the preset you used by its plain name, and what you decided on their behalf so they can correct it. Offer one variation.

Never repeat a preset's internal guidance field back to the person, never name a model, and never discuss what anything cost.

## When to stop or decline

- Nothing in the catalogue fits: say so and offer to make it without a preset.
- A preset needs source material the person does not have: say exactly what is missing before you spend anything.
- The person wants a real, named public figure in the result: decline.

View saved version →

---
name: voiceover-for-a-script
description: Use this when someone wants spoken audio, such as a voiceover, narration, a read of a script, an announcer line, background music or a sound effect. Trigger on voiceover, narrate this, read this out, VO, announcer, background track, sound effect. Do not use it to make video, and do not use it to clone the voice of a real person.
---

# Voiceover for a script

Turn written words into a finished audio file, or make the music and effects that sit under a clip.

## When to trigger

Trigger on "record this", "narrate this", "I need a voiceover", "give me an announcer read", "put some music under it", "I need a whoosh".

Do not trigger when the person wants a talking character on screen. That is video, so use make-a-video.

## Ask for this first

Ask once, bundled:

1. The exact words to be spoken. If they have not written them, offer to draft the script first and get it approved before you record anything.
2. The voice: warm, bright, deep, calm, energetic, the age range, and the accent.
3. The pace and roughly how long it should run.
4. What it sits under, so the read matches, for example a product ad, a documentary, a channel intro.
5. For music or effects instead of speech, what the scene is and how long the bed needs to be.

## Steps

1. Call `nolgia_list_models` and look at the audio models, so you pick one that covers speech, or music, or effects, whichever this is.
2. Call `nolgia_text_to_audio`. For a read, put the exact approved words in `prompt` and describe the delivery in `voice`. For music or an effect, describe the sound in `prompt` and set `duration_seconds` to the length the scene needs. Set `format` if they asked for a particular file type.
3. Audio comes straight back. Call `nolgia_get_asset` if you need a fresh link later.
4. If this is going under a clip you already made, call `nolgia_add_project_assets` so the read and the picture live in the same project.

## What to return

The audio file, the exact words as recorded, and the length. Offer one alternative read at a different pace or in a different voice, since the first take is rarely the keeper.

Never name a model and never discuss what anything cost.

## When to stop or decline

- The script is not approved: get the words right first. Re-recording is wasteful and annoying.
- The person asks for the voice of a named real person, living or dead: decline, and offer to describe the qualities they are after instead.
- The words are a claim the person cannot make, such as a medical or financial promise: flag it before you record, not after.