A desktop AI conversation that thinks, types and streams — pulsing dots, blinking caret, rotating model picker and action icons. Yours to script, turn by turn. Pure code. Free, in your browser.
CONVERSATION
MODELS
ASPECT & TYPE
ANIMATION
COLORS
EXPORT
THE AI THINKS (PULSING DOTS), TYPES THE REASONING, THEN STREAMS ITS REPLY WORD BY WORD WITH A BLINKING CARET — MODEL PICKER ROTATES YOUR MODELS BETWEEN TURNS. PURE CODE, NO CANVAS.
Half the internet's storytelling now starts with a screenshot of a chat, and the AI chat tool lets you produce exactly that kind of screen fiction with full control over every word. The tool renders a desktop AI conversation: your prompt pops in as a user bubble, the model thinks with pulsing dots, optional reasoning appears in a quiet italic line, and the reply streams word by word with a blinking caret before action icons fade in underneath like the model just finished being helpful.
Nothing here connects to a real model, which is the point. You write every turn, so the conversation says exactly what your video needs, at exactly the pacing your edit wants, and no real product's interface ends up in your footage with its real estate of logos. The tool runs in your browser, renders every frame on your device and exports the finished scene for your timeline, because a fake interview should not require a real subpoena.
The renderer draws a desktop style chat card with a model picker at the top, a scrolling conversation area and an input bar at the bottom. Each turn you script has three parts: the user message, optional reasoning text and the AI reply. When the animation plays, the user message slides in, the dots pulse for as long as you configured the thinking time, the reasoning types out if it exists, and then the reply streams word by word at the words-per-second you picked, closing with a copy row and a blinking caret that disappears on schedule.
Because pacing is a first class control, the same conversation can feel like a quick desk lookup or a late night deep thought session purely by moving the think and typing sliders. Four aspect ratios cover every placement: 16:9 for fullscreen scenes, 9:16 for reels, 1:1 for square cards and 3:4 for feed shaped stories.
Comedy sketches use conversations that escalate politely into nonsense. Explainer channels open with a question a viewer would genuinely search, and let the fake reply set up the real information. Product marketers script a fictional assistant praising their launch with the confidence of a keynote. Nobody has to know the model did an unpaid internship in Improvisation, because the interface looks like the real thing and the words are yours.
There is also a privacy engine here that screenshots cannot offer: no real account names, no real contact photos and no risk of exposing a private thread while trying to dramatize one. Clients and at least one category of paranoid cats would prefer it this way.
Script conversations as setups and payoffs. The first turn establishes the world, a strange but plausible request typed by a human with places to be. The second turn sharpens or ruins everything, and the third exists only if the joke climbs. Because each turn animates sequentially with pauses you control, you can cut away from the chat mid stream and return later, letting the conversation function as a recurring character in the video instead of one uninterrupted block.
Batch by mood: render a light theme daytime conversation and a dark theme late night conversation in one sitting, and your channel owns two distinct fictional laptops forever.
Fonts load locally at preview time, Inter and Space Grotesk sit beside a system option, since fake software deserves real typography. The model picker is fully scriptable, so Lumina 3 can be replaced by any invented name your legal department and your comedian can both live with. Text size scales the whole card proportionally, so the fiction remains legible at both 16:9 and 3:4 without re-typesetting anything.
The animation timing is honest math: thinking time, letter streams and inter turn pauses are all in milliseconds, which is why two exports of the same script can feel like completely different conversations depending on patience settings.
Beside the video export, the tool can download a standalone HTML file: a self running page that plays the entire conversation on loop at any window size. Open it fullscreen on a laptop or a second monitor and screen record in one take, which is the cleanest way to capture long conversations at any resolution without waiting for a video render. The file plays offline, carries no watermark, and loops until you close the tab, which makes it ideal for backgrounds on streams as well as for edits.
Script the chat in the same pass you write your hook, because the fake conversation IS a hook format: one specific prompt and one confident reply replace thirty seconds of throat clearing. In edits, cut away mid stream and return later; the streaming animation is forgiving to interruption in a way real apps are not, which is one more thing fiction does better than reality.
Theme workflows are a productivity trick: render the same script in light and dark once, and your channel owns a daytime assistant and a night assistant with zero extra writing.
The timeline math is exposed: thinking time is milliseconds, streams run at words per second, and pauses between turns are separate, so two exports of one script can feel like completely different personalities. The reasoning line exists because watching a machine hesitate, in italics, is a genre all by itself.
Fonts load locally at preview time and nothing about your script leaves the tab, which keeps the fake assistant between you and your conscience, where all the best fiction lives.
From empty script to exported conversation.
No mystery sliders, one line each.
Small things, big difference.
Answered the way a colleague would answer them.
The refinements that separate a montage from a moment.
Answered the way a colleague would answer them.
The AI chat tool is fake code delivered with real craft: every pixel drawn in your browser, every word scripted by you, every export clean. Script a conversation only your audience would believe and let the studio handle the interface.