One clip and one paragraph of text. That's all it takes. Arcads has Google's Omni Flash built in, so you simply describe how your video should look. Cuts, background, colours and text pop-ups follow from that. Your face and your voice stay as they were, and so does the original audio.
Last updated:
A Claude account, an Arcads account and a finished clip. A talking-head recording works best: you, facing the camera, at a microphone. You don't need editing software, and you never see a timeline.
All three prompts describe the microphone: a black desk mic with a green button and a stand. That's the setup the prompts were written for. Put your own microphone in there, or the model will look for something that isn't in your shot.
Arcads then shows up in your list of connectors and is available in every new conversation.
Attach your clip, paste one of the three prompts and tell Claude to use Google's Omni Flash. The model handles the cuts, the background, the motion graphics and the text pop-ups.
All three keep you photorealistic and leave sound and voice untouched. They only differ in look, so you can run them one after the other on the same clip and pick the one you like best.
Edit this uploaded video and keep my face, body proportions, skin texture, clothing, movement, timing, gestures, camera framing, original voice, and original audio exactly the same. Keep me photorealistic and unchanged. Do not turn me into a cartoon, a drawing, or an animated character. Preserve the microphone (black desk mic with a green button and stand) as the main foreground subject, position clearly visible throughout, do not remove, hide, replace, or change it. Restyle the whole clip as a bold mixed-media collage with a stop-motion feel: cut me and the microphone out with a rough paper-cutout edge and a subtle white outline; replace the background with layered crumpled paper, torn newspaper, halftone dots, coloured tape and bold flat paper shapes; use a strong palette of red, acid yellow, black, off-white and small blue accents; add energetic hand-drawn arrows, circles, scribbles and sticker-style elements around me. Add EXACTLY these three text pop-ups, once each, in order, and no other text at all: 1) "GOOGLE KILLED CAPCUT" 2) "100% AI" 3) "GOOGLE OMNI". Do not transcribe speech or add captions. Slightly choppy stop-motion energy, subtle camera shake, light glitch, but keep my face and the mic clear. Bold, playful, striking.
Edit this uploaded video and keep my face, body proportions, skin texture, clothing, movement, timing, gestures, camera framing, original voice, and original audio exactly the same. Keep me photorealistic and unchanged, no cartoon, drawing, or animation. Preserve the microphone (black desk mic with a green button and stand) as the foreground subject, same position, visible throughout, do not remove, hide, replace, or alter it. Restyle as a sleek modern tech aesthetic: a clean dark background with soft gradient lighting (deep charcoal into warm amber), subtle glassmorphism panels, thin neon accent lines, a gentle depth-of-field glow, and me cut out with a soft rim light. Add minimal kinetic motion graphics: thin animated underlines, small floating UI cards, smooth directional swipes, sparse, never cluttered. Add EXACTLY these three text pop-ups, once each, in order, and no other text at all: 1) "100% AI" 2) "GOOGLE OMNI" 3) "NO CAPCUT". Do not transcribe speech or add captions. Keep my face and the mic clear. Premium, clean, striking.
Edit this uploaded video and keep my face, body proportions, skin texture, clothing, movement, timing, gestures, original voice, and original audio exactly the same. Keep me photorealistic and unchanged, no cartoon, drawing, or animation. Preserve the microphone (black desk mic, green button, stand) in the foreground, same position, visible throughout, never remove, hide, replace, or alter it. Restyle as a cinematic, film-graded look: a rich teal-and-orange grade, soft cinematic bloom, gentle film grain, moody key lighting. Introduce dynamic angle changes and subtle push-ins and pull-outs on the cuts, as if shot on multiple cameras, while keeping my face centred, sharp, and readable in every angle, and the mic clearly in frame. Add restrained motion graphics: clean lower-third accent bars and smooth light streaks on transitions only. Add EXACTLY these three text pop-ups, once each, in order, and no other text at all: 1) "EDITED BY AI" 2) "GOOGLE OMNI" 3) "ONE PROMPT". Do not transcribe speech or add captions. Never cover my face or hide the mic. Cinematic, bold, striking.
Each of the three repeats the same rules: keep the face unchanged, keep the microphone in the shot, exactly three text pop-ups and nothing else. It looks overdone, but it's the reason you end up with a clean clip.
Take the strictness out and the model starts inventing subtitles and beautifying your face. The repetition looks clumsy, but it's what stops the model from drifting.
You turn into a cartoon character. The line with "no cartoon, drawing, or animation" got lost. Put it back in, right before the line that starts with “Restyle”.
Subtitles suddenly appear. That happens as soon as "Do not transcribe speech or add captions" is missing. The sentence belongs in every prompt, including your own.
The microphone disappears or gets replaced. Describe it more precisely: colour, shape, where it stands. The more specific you are, the more likely it stays.
I picked up this trick from Cindy Zhu. The three style prompts are hers and appear here unchanged, exactly as she tested them. The guide around them is mine.
In a workshop we build this with your team on a real case from your daily work. You leave with a workflow your team can repeat on the next video.
Discuss your workshop use case