Lore Clip — AI Roleplay Chat to 9:16 Video, No Watermark

You wrote a person. What you can actually see is a wall of text.

The room, the light, the way they moved — it is all in there, in paragraphs, inside one browser tab. Lore Clip takes one stretch of it out and makes it move.

✦

The material is the story you already played out

Nothing to re-write, nothing to go find. Whichever stretch you pick is the one that gets filmed — the place, the action, the light all come out of those messages.

◈

The character you know, moving on screen

The render carries that character's own reference artwork, so the person moving is the one you have been talking to — not a stranger who happens to resemble them.

▸

A vertical clip you can post as it is

9:16, no watermark, with an audio track — and if the screenplay found one line long enough, that line is spoken aloud. It is already the shape a short-video feed wants, so your character can be seen by people who will never read the chat.

Using it: from one chat page to one clip, in four steps

Step 3 goes online for certain, and step 1 sometimes does — only when the page it is on ships no artwork of its own. Choosing the stretch always happens in the side panel, in front of you.

Flow diagram: chat window on the left, processing in the middle, finished vertical video on the right

1

Open the side panel on the chat page

The panel opens beside the page and reads the messages already on screen. On sites where a signed-in page ships no artwork, it loads that same URL once more, purely to recover the character art — the video model needs a reference image. On the others that second request never happens.

2

Pick the stretch to film

Tick the messages you want, on the chat page itself. A checkbox appears left of every message; tick one, shift-click another to take the run between them, and press Esc to clear the lot. A Latest 20 button in the panel takes the newest handful in one click. There is a cap of 20 messages per submission — tick more and the extras are left out, which the panel says rather than doing quietly.

What you tick is where the clip draws from, not a shot-by-shot plan: the panel reports how many messages this page holds, how many you ticked, how many seconds of material that is, how much of it is narration, and whether the character art came through. If every message you ticked is dialogue, the material line turns red — narration is what the camera can actually show, so a selection with none of it leaves the model to invent the whole frame. You can still go ahead.

3

Submit

The messages you ticked go out, along with the public image URL of the character art. Where the chat page opens with a paragraph the platform itself wrote about the character, that one block goes too, ticked or not — it is what the model needs to know who she is, not just what happened in the stretch you picked. The panel names it, and its length, before you press anything. On the server a text model reads all of it and writes a screenplay — what the shot holds, where the scene is, what the light is doing — and that screenplay is what the video model is handed.

About 3 minutes of waiting

4

Download it, then post it

A 9:16 mp4, with no watermark. It arrives with an audio track — the video model generates sound along with the picture — and if the screenplay found one line long enough to fill the shot, that line is spoken aloud and burned across the lower third. Save it and put it wherever your character should be seen.

While it renders, and what comes back

The wait is the one part you cannot watch closely. Here is what the panel is doing with it, and what the other end of it looks like.

▸

Five stages, and no countdown

The panel names the stage it is on: fetching the art, writing the screenplay, building the prompt, rendering, then fetching the clip. It shows how many seconds have passed. It does not show a percentage or a countdown — there is no honest number to put there, because almost all of the time is spent waiting on the video model.

◈

Close the panel; it keeps going

The render runs on the server, not in the tab. Close the side panel, close the browser, come back later — the job carries on, and the next time you open the panel it picks the job back up where it was. If the panel is still open when the clip lands, it sends a desktop notification. The notification carries the length and the wait, and no text from your chat.

✦

What lands, and how long the link lasts

A 9:16 mp4 called loreclip-<id>.mp4, no watermark. It plays muted and on loop in the panel; press Sound on to hear the audio track. The download and preview link is signed and expires after 24 hours — if it goes stale, the panel says so and a Refresh link button issues a fresh one. The file you saved is yours to keep.

!

When it does not go

A page that could not be read says so and offers to read it again. A page with no character art leaves the button disabled, with the reason spelled out — the render needs a reference image. Not enough credits points at the top-up screen. And a render that fails costs nothing: the charge is taken only on a clip that came back.

Questions

Starting with the four things worth knowing before you install anything. Every line is the current state of things, not marketing caution.

! One line at most — you can't read the conversation This is the big one. The screenplay looks through the messages you ticked for a single line long enough to fill the shot: if it finds one, your character says it out loud and that same line is burned across the lower third. Everything else is picture only — every other line, every piece of narration — and if no line is long enough, the clip has no words in it anywhere. The words are painted by the video model along with the rest of the frame, so where they sit and where they break are out of our hands, and a word is occasionally misspelled. What you get instead is the picture itself — the action, the light, the place the narration was describing, which is most of a roleplay log anyway.
! Only three chat platforms are supported More are next, and they ship as configuration rather than as code. But today the list is three sites: PolyBuzz, OurDream and Character.AI. On each of them the field mapping was measured against its actual pages, so extraction is exact rather than a best-effort guess.
! One clip takes about three minutes Almost all of that is waiting on the video model. This is not an instant tool, so don't come in expecting one. You don't have to sit and watch it, though. Submit, close the panel, go do something else — the job keeps running on the server, and the panel picks it back up the next time you open it.
! Clips are short, and each is a single continuous take A dozen-odd seconds at most, with no cuts and no change of location. If you want multi-scene editing, this is not that tool. One continuous take is also what keeps the character recognisable: with no cut, there is no second shot for them to come back from looking like someone else.

Can I install it yet?

Yes. The extension is live on the Chrome Web Store. Click the install button on this page and it takes you straight there. Desktop Chrome only, three platforms supported: PolyBuzz, OurDream and Character.AI.

Which chat platforms does it work on?

Three: PolyBuzz, OurDream and Character.AI. That is the whole list. Supporting a platform means its page fields were measured one by one, and adding the next one ships as a configuration change rather than a new version of the extension, so the list can grow without you reinstalling anything.

Each name links to a page about what it looks like on that platform — which picture it uses, and what is different there.

Lore Clip is not affiliated with any of them. The names are here so you can tell whether this is any use to you.

Are there captions? Does anyone speak?

One line, at most. The screenplay looks through the messages you ticked for a single line long enough to fill the shot. If it finds one, your character says it out loud and that same line is burned across the lower third of the picture. If it doesn't, the clip has no words in it anywhere and the audio track is ambience generated to match the picture.

Nothing else gets words: every other line, and every piece of narration, is picture only. The words are painted by the video model along with the rest of the frame, so where they sit and where they break are out of our hands, and a word is occasionally misspelled.

Why does it take three minutes?

Almost all of it is waiting on the video model, and that part doesn't get faster. Picking the cut and writing the screenplay take a few seconds.

Is there a watermark? Any limit on where I can use the clip?

No watermark. The clip is yours.

It does come with sound: the video model generates an audio track alongside the picture. If the screenplay found a line long enough to fill the shot, that line is spoken aloud and the same line is burned across the lower third. If it didn't, the track is sound generated to match the picture, with no words in it.

Do I need to sign in?

There is a Google sign-in, used to keep your render history and usage. It runs in Chrome's own sign-in window; the extension never touches your Google password, and the authorization code is exchanged server-side using the server's own credentials.

What will it cost?

Pay as you go, by the second of finished video: 1 credit buys 1 second, so a ten-second clip costs 10 credits. Signing in gives you 10 credits, which is one clip, and after that credits come in one-time top-ups of $4.90 for 30, $8.90 for 60 and $21.90 for 180. There is no subscription, credits do not expire, and a render that fails costs nothing.

The full table is on the pricing page.

How is this different from the video feature the chat platform already has?

Two things: the output has no watermark, and you choose which stretch becomes a clip instead of the platform choosing for you.

How do I choose which messages to film?

Follow the narration, not the dialogue. Narration is what the camera can actually show — the place, the action, the light — so a stretch whose narration describes something visible films far better than a stretch that is only people talking. Around ten seconds of that is the target, and the panel counts it for you as you tick.

You do not have to be precise. What you tick is the material the model draws on, not a shot-by-shot plan: it reads those messages and decides what the frame holds. Picking a coherent little scene works better than picking scattered highlights. If the material line turns red, everything you ticked is dialogue and there is nothing describing the picture — add a narrated message or two.

What if the render fails, or the page cannot be read?

A page that could not be read says so, with a button to read it again — usually the chat page just needs a refresh. A page with no character art leaves the generate button disabled and explains why: the render needs one reference image of the character, and there is no text-only path. Not enough credits points you at the top-up screen, which opens in a separate tab and takes a card on Stripe's own page.

A render that fails costs nothing. Credits are charged on the clip that came back, so a job that died on the video model leaves your balance alone. A failed job also does not block the next one: submit again whenever you like.

Let more people meet your character.

The character who has only ever been text can be moving on screen tonight — in a vertical clip you post yourself.

Install Lore Clip Read what it can't do yet

Your character is still only text. Now let them move.