Freestyle: Type It, See It, Instantly

What it does

Face swap, outfit swap, and style morph are all fast because they're built on presets. Freestyle removes the presets entirely: describe anything in plain text — "eyes glowing blue," "hands on fire," "turn the room into a snowstorm" — or upload your own reference image, and LiveGen applies it live on your camera feed. No waiting for a render, no fixed menu of looks.

This is the same real-time engine behind every other LiveGen mode, opened up to custom input — so if you can describe it or show a reference image for it, you can try it live. Where the preset modes answer "make me an anime character" or "put me in a new outfit," Freestyle answers "make me whatever I can type." That makes it the mode you reach for when your idea is too specific, too weird, or too new to have its own button yet.

How it works

Under the hood, your camera feed streams frame by frame over WebRTC to the Xmax X2.0 model. The model reads your AI video freestyle prompt (and any reference image), interprets what you asked for, and paints the result back into a standard <video> element on the page — continuously, while you move. There's no upload step, no job queue, and no export-and-download loop between typing an idea and seeing it. For the background on how live AI video generation like this actually works, see real-time generative video.

What you can do with it

• ✓Trigger custom effects on stream — text-command style transformations that respond to what your chat is asking for, similar to the "eyes shoot lasers" and "flame in palm" effects popularized by X2.0 live demos. Because you're writing a custom prompt live, you can invent a new effect mid-broadcast instead of being stuck with a fixed effect list. Pair it with AI video effects for TikTok for short-form content that doesn't look like everyone else's filter.
• ✓Prototype a look before it becomes a preset — test an idea instantly instead of waiting on a scripted feature. If a Freestyle prompt keeps working, it's a candidate for a saved look.
• ✓Combine a reference image with a text prompt for more precise control than text alone — the image sets the overall style or character, and the text nudges the details.
• ✓React to a live audience — take a suggestion from chat, type it, and show the result in seconds. This is where real-time AI video effects beat pre-rendered content: the payoff happens on camera, live. That makes it a natural fit for face swap for live streaming sessions where the audience drives the show.
• ✓Cover the gaps between preset modes — need something that's part Face Swap, part Style Morph? Describe the combination in one prompt instead of choosing between two buttons.
• ✓Bring an on-brand or original character to life — upload your own art as the reference image and drive it with text, without needing it added to a preset gallery first.

When you want a quick, structured effect without writing anything, the dedicated real-time video filter tool covers the common looks — and you can switch to Freestyle the moment you need something the presets don't have.

Tips for the best results

• ✓Be concrete, not clever. "Glowing blue cybernetic eyes" beats "make me look futuristic." Name the object, the color, and the material you want changed.
• ✓Change one thing at a time. If the result drifts, remove a clause instead of piling on more words. Short, specific prompts are easier for the model to hold steady frame to frame.
• ✓Use a reference image for identity or style you can't easily describe. A particular art style, a specific character, or an exact color palette is often faster to show than to spell out — and you can still layer a text prompt on top.
• ✓Keep good, even lighting. Consistent front lighting gives the model a stable subject to work from, which keeps a real-time effect from flickering as you move.
• ✓Iterate live. The whole point of a real-time AI video freestyle prompt is the fast loop — tweak a word, watch what changes, keep what works. Treat the prompt box like a live dial, not a submit button.
• ✓Mind consent and content rules. Prompts that recreate a real person's likeness without their consent, or that request restricted content, aren't allowed — LiveGen moderates uploads and applies its content policy.

Real-time, not render-and-wait

Most "AI video" tools are batch generators: you submit a prompt, wait in a queue, and download a clip you can't change without starting over. Freestyle works the opposite way. Your AI video freestyle prompt applies to the live feed as you type it, so editing the idea and editing the video become the same action. That changes what the tool is good for — instead of planning a clip in advance, you improvise on camera, respond to a live audience, and keep the takes that land. It's live AI video generation you steer in the moment, not a render you wait on.

Because everything happens in the browser over WebRTC, there's nothing to install and nothing to download when you're done — the result is already live, and it's ready to share the instant it appears.

Why LiveGen

• ✓Real-time, not a queued render. Your prompt applies live — no submitting a job and waiting for a result to come back.
• ✓Runs in your browser. No local AI setup, no model download, no command line — just a camera, a text box, and a browser tab, on desktop or mobile.
• ✓One engine, every mode. Freestyle shares the same real-time core as Outfit Swap, Bring to Life, and Summon, so custom prompts behave consistently with the presets you already know.
• ✓Instantly shareable. Whatever you create gets a shareable link the moment it happens, ready to post without exporting. (Free-tier exports are watermarked; paid tiers remove the watermark and add HD — verify current pricing.)