--- title: "The Birthday Champagne AI Video" description: "That viral champagne-and-fire clip - with your own face, your outfit, your city. The exact tools, prompts, and fixes, nothing left out." author: Agentmatik (https://agentmatik.ai) source: https://agentmatik.ai/guides/birthday-champagne-ai-video date: "5 July 2026" tools: "higgsfield,nano-banana,seedance" --- ![The final video](/guides/champagne/final-video-v2.mp4) ## What's inside - The one ordering trick that saves you real money on credits - The photo workflow: background, you, small fixes - with every prompt - The 10-second video prompt that makes it look real, not AI - The 6 details that make people stop scrolling (and what breaks the illusion) Made with Nano Banana Pro (the photo) and Seedance 2.0 (the video). About 30-60 minutes. No editing skills needed. ## The one thing that saves you money Most people jump straight to the video. That is where the money goes - making videos burns through credits fast. The people whose clips actually look real do the opposite: get one great photo first, then turn it into the video. Get the photo right and the video almost makes itself. So: **the photo is the real work. The video is the easy part.** ## The whole thing in 5 steps - **1.** Get the tools (Higgsfield is the easy one). - **2.** Add 3 free Claude skills so your prompts come out great. - **3.** Make the photo in Nano Banana Pro - background, then you, then small fixes. - **4.** Turn it into a video in Seedance 2.0 with one prompt. - **5.** Post it. About an afternoon. A few dollars of credits - if you get the photo right first. ## What you need **Higgsfield** if you are starting out - easy to use, and every model you need is in one place. [Kie.ai](https://kie.ai) if you want to run it from Claude Code - more powerful, but you have less control over spending, and video eats credits fast. Two AI models do the work: Nano Banana Pro makes the photo, and Seedance 2.0 turns it into the video. And 3 free Claude skills make your prompts much better - they teach the AI how to prompt these exact tools: [banana](https://github.com/AgriciDaniel/banana-claude) (photo prompts), [seedance-prompt-en](https://github.com/dexhunter/seedance2-skill) (video prompts), [higgsfield-generate](https://github.com/higgsfield-ai/skills) (images and videos in one step). ## Step 1 - The photo (this is 80% of it) One photo of you sitting down - cake on one side, champagne on the other. This becomes the first frame of your video, so it has to look like you and be set up right. ### A) Make the background Pick a spot people will recognize - I used a hill over Prague with the TV tower in the view. Make it in Nano Banana Pro at 9:16, with your real photo of the spot attached. ![Real photo of the Prague lookout](/guides/champagne/real-spot-v2.jpg) ![AI-generated night bench scene](/guides/champagne/generated-bench-v2.jpg) ```prompt Background prompt A quiet empty wooden park bench viewed straight on, centered in the frame, sitting on a brick-paver path with a few scattered autumn leaves in the foreground. Behind the bench the ground drops away to a panoramic night view of Prague: glowing rooftops and warm window lights, and on the right the illuminated Zizkov television tower lit in red, white and blue with a single vertical blue light beam rising straight up into a deep blue dusk sky. Bare tree branches frame the top corners and dark foliage frames both sides. Empty, still, no people. Composition: symmetrical, eye-level, the whole bench visible from armrest to armrest with generous foreground path and headroom above. Shot on a full-frame camera at blue hour, soft ambient city light, gentle warm bokeh from distant streetlamps on the left, natural low-light look. Vertical 9:16. No text, no watermark. ``` ### B) Put yourself in the photo This is the step that makes or breaks it. Attach the background plus 3-4 clear photos of your face (front, side, angle) and your outfit. The thing most people get wrong: keep your face big in the photo (waist-up, not full-body), and use a few clear face photos - not one tiny collage. If your face is small, the AI invents a stranger. Tell it plainly: "a real, specific person - do not make him younger, slimmer or better-looking." ![Empty AI bench scene](/guides/champagne/generated-bench-v2.jpg) ![Matt placed into the scene](/guides/champagne/matt-in-frame-v2.jpg) ```prompt The prompt people usually share online Create a photorealistic birthday luxury lifestyle / night editorial photo. In the frame, a beautiful girl sits on a wooden bench outside at night. She is wearing an elegant black tight midi dress with thin straps and black strappy high heels. Her legs are beautifully crossed, the pose is confident, feminine, and slightly cheeky. The girl has voluminous hair, soft glamorous makeup, glowing skin, nude lips, and a calm, confident look into the camera. In one hand, she holds a bottle of champagne from which foam and splashes are spectacularly flying out. In the other hand, there is a small birthday cake on a stand with burning candles. Location: a night city street or park, dark background, soft lights in the distance. Light: flash like in expensive film photography, bright accent on the face, body, cake, and champagne splashes, the background is darkened. Style: photorealistic, high-end editorial, ultra realistic, night flash photo, detailed skin texture, cinematic contrast. Important: realistic anatomy, natural hands and fingers, no facial distortions, no extra objects, no text, no watermarks, no cartoonish look. ``` ```prompt My photo prompt (the one I used) Photorealistic night photograph, vertical 9:16, MEDIUM shot framed from the knees up (the man's face is large in frame for maximum likeness). A man sits centered on the wooden park bench at night, with the city skyline, the illuminated red-white-blue tower and its vertical light beam behind him (match the attached bench scene). IDENTITY - render his face and head to EXACTLY match the FACE reference photos: same face shape, eyes, nose, thick black rounded glasses (always worn), full dark beard with light-grey flecks, short dark hair greying at the temples, fair skin, solid build. He is a real specific person - do NOT idealize, slim, or de-age him; he must be clearly recognizable. Grayscale tattoo sleeve on the LEFT forearm. WARDROBE - he wears a white tuxedo jacket over a crisp white dress shirt with a black bow tie (take the outfit from the tuxedo reference photo). Tailored, elegant. He has a calm, cool expression - a neutral POKER FACE or a very slight closed-mouth smile - looking straight at the camera. In his LEFT hand (left of frame) a small chocolate cake topped with a SINGLE gold metallic numeral-"5" candle, one lit flame, no other candles. In his RIGHT hand (right of frame) an upright dark champagne bottle. Realistic skin texture, natural anatomy and correct hands, cinematic night flash photography, high detail. Resolution 2K-4K. No text, no watermark. ``` ### C) Small fixes Two quick edits help a lot: lift the cake up so the fire has room to spread over it, and (optional) empty the bench if you want to walk into the shot. I skipped the walk-in - it looked too fake, and I would rather use the 10 seconds on the big moment. ![Before: cake held low](/guides/champagne/matt-in-frame-v2.jpg) ![After: cake lifted](/guides/champagne/cake-lifted-v2.jpg) ```prompt Lift the cake Edit this photo: the man RAISES the chocolate cake (with the gold "5" candle) higher and extends his LEFT arm outward to his left, holding the cake up around chest/shoulder height, away from his body and clearly out to the LEFT side of the frame. Keep the champagne bottle held out to his RIGHT. Keep everything else identical - same face, glasses, beard, white tuxedo, bench, skyline, tower beam, lighting and framing. Photorealistic, vertical 9:16, no text, no watermark. ``` [[newsletter]] ## Step 2 - The video (the easy part) Drop your photo in as the first frame and describe what happens, step by step. Settings: 10 seconds · 9:16 · 1080p · high quality. ```prompt My video prompt (the one I used) @image_1 as the first frame. Keep his exact face, glasses, beard and white tuxedo identical every frame. NO last frame - the clip ends on the action. Vertical 9:16, one continuous take, photorealistic night. Prague hilltop bench; behind him the skyline and the illuminated red-white-blue tower with a vertical light beam. CAMERA: stays at a MEDIUM waist-up framing the whole clip - only the faintest slow drift in. NEVER a tight close-up (that breaks the likeness). Keep his full upper body, the cake and the bottle in frame the whole time. 0-1.5s (real time): He sits with a cool, neutral deadpan POKER FACE - chocolate cake with the gold "5" candle raised up and out to his LEFT at chest height, upright champagne bottle held LOW in his RIGHT hand with its base on the bench seat. The "5" flame flickers. 1.5-2.5s (real time): He firmly PUSHES the bottle base straight DOWN onto the bench seat. The CORK visibly bursts off the top and shoots HIGH up into the frame on the RIGHT - spinning against the dark sky with a puff of vapour, clearly seen. Give the cork its own clear moment. 2.5-3s: Speed ramps down into dramatic SLOW MOTION as champagne begins to surge from the bottle. 3-7.5s (extreme slow motion, almost a freeze at the peak): RIGHT - a tall VERTICAL champagne fountain shoots straight up, foam and droplets glinting in the air. LEFT - a big orange-yellow fireball ignites AT the "5" candle and blooms ACROSS the cake, with rolling flame, dark smoke and rising embers. His head tilts DOWN toward the cake, eyes closed mid-blow, face lit between the two effects. The two effects stay on opposite sides and NEVER cross the centre. 7.5-8.5s (eases back to real time): Fire fades to drifting smoke on the LEFT; champagne weakens to fizzing drips on the RIGHT. 8.5-10s (ending): He holds the deadpan, then breaks into a faint, confident SMIRK - cool and unbothered - as smoke drifts across the tower beam. Continuity: cake with the "5" stays raised in his LEFT hand on the LEFT; bottle base stays on the bench on the RIGHT - never swap or twist. Deadpan throughout (only the blow, then the smirk). Fire starts at the cake and goes away from his face; fire and champagne never mix or cross the centre. Realistic fire and liquid physics only - no sparkles, fireworks, confetti, text, captions or watermark. Audio: night city ambience and wind; a wooden bottle-tap, then a loud cork POP and explosive champagne fizz on the right; a fire WHOOMP and crackle on the left. No speech, no music. ``` ## What makes it look great - **Keep a straight face.** Staying calm while everything explodes around you is the joke. - **Keep the two effects apart.** Champagne on one side, fire on the other - a messy middle looks fake. - **Show the cork fly.** Most clips hide it. Let it pop and shoot up - that is the "wait, how?" moment. - **Slow it right down** at the big moment, then speed back up. - **Don't zoom in too close.** A tight close-up stops looking like you. - **Your first photo is your thumbnail.** Make it genuinely good. ## Two quick pro tips **1) Tag your photos.** In Higgsfield you can tag an image in the prompt so it is clear which one you mean. Right-click an image to set it as the start frame, end frame, or just a reference. **2) Connect Higgsfield to Claude** so you can prompt it straight from chat: Claude → Settings → Connectors → Add Custom Connector → name it Higgsfield, MCP URL https://mcp.higgsfield.ai/mcp → Connect and sign in. No API key needed. ## Before you start - Higgsfield or Kie.ai account ready - 3 skills added: banana, seedance-prompt-en, higgsfield-generate - Background made - Photo of you - looks unmistakably like you - Cake lifted / small fixes done - 10-second video made - Posted - and tagged me Questions, or made one? Tag [@mattzimak.ai](https://www.instagram.com/mattzimak.ai) - I share builds like this whenever there is one worth sharing.