I run a small cleaning business, and I finally cracked the before and after content problem. You know the one. You finish a job, pack up, and only then realize you never took the "after" from the same angle as the "before." Half the time I forgot the before shot too. So my best results never made it online.
Then I tried a Zoviz Canvas pipeline, and it fixed the whole thing from one photo. I dropped in a single "before" shot of a genuinely chaotic child's bedroom, and it gave me four things ready to post: an honest AI "after" visualization, a labeled split screen comparison, a satisfying morph video that transforms the room, and a full caption kit. I ran it under my company name, FreshNest Cleaning, and I want to show you exactly what came out, the prompts behind every step, and how you can do the same.
Try it live: open the Before-After Transformation Ad canvas, hit "Clone to edit," and swap in your own photo.
What this canvas produces from one photo
The entire pipeline starts with a single upload. In our test, this was the before photo, and next to it, the AI generated after:
1. The AI "after" visualization
The same room, same camera angle, same window light, same furniture, but showing the realistic result of a professional deep clean. The bed is made, the floor is clear, the wardrobe is closed, and nothing structural has been invented.
2. A labeled split screen comparison
A vertical 9:16 image with the before on the left, the after on the right, a thin white divider, and small BEFORE and AFTER labels. This is the classic before and after picture format that works as a TikTok cover, an Instagram Story, or a website banner.
3. A before and after morph video
The video literally starts on the real before photo and ends on the AI after image. Dirt lifts away, surfaces brighten, and order appears, like a time lapse of the service being completed. The camera stays locked, there are no cuts, and the pacing is calm and satisfying, the exact style that performs on CleaningTok.
4. A posting kit
A second LLM node reads the brief and both images, then writes the caption, hashtags, and hook. This is the actual output from our run:
CAPTION:
Before vs. after: this bedroom got the full FreshNest treatment 🧹✨ Toys, papers, clothes everywhere… and now? Room to breathe. We deep cleaned, decluttered, organized, made the bed with fresh linens, and vacuumed the whole space. It's the same room, just cared for. Book your deep clean via the link in bio 🔗
HASHTAGS:
#CleaningTikTok #BeforeAndAfter #DeepClean #RoomTransformation #FreshNest #SatisfyingClean #TidyHome #HomeOrganization #CleanHome #BookNow
HOOK (first 3 seconds on screen):
"We need to talk about this bedroom… 😬"Posting this on Instagram or TikTok? A strong before and after clip still needs the right posting habits around it. Our guide to boosting Instagram engagement covers timing, captions, and format choices that help transformation content actually get seen.
How the pipeline is wired
Zoviz Canvas is a visual AI workflow builder: you connect prompt notes, images, LLM nodes, image generation nodes, and video generation nodes on an infinite board, and each node feeds the next. This particular canvas uses nine working nodes plus a pinned instruction note.
The IMAGE node holds your before photo. It fans out to almost everything downstream, the Director LLM, both image generation nodes, and the video node all reference it, which is what keeps every output anchored to the real scene.
The SERVICE BRIEF prompt note is the only text you need to edit. It describes, in plain language, what your business actually did. Ours read:
Prompt · service brief
SERVICE BRIEF: Service: deep cleaning and tidying. Business: FreshNest Cleaning. What you did: full room deep clean - decluttered all surfaces and the floor, made the bed with fresh linens, organized loose items into their places, vacuumed the carpet, wiped and dusted all surfaces. Where: bedroom. Tone: satisfying. Platform: TikTok. CTA: book via link in bio.The Director LLM node is the brain of the canvas. It receives the brief and the before photo and writes one perfect image generation prompt for the after shot. Its system prompt keeps it honest:
Director LLM · system prompt
You are a transformation visualizer for service businesses. Your entire reply is always ONE single paragraph image generation prompt, nothing else. Never change the scene beyond what the described service would realistically change. The image must contain NO text, no words, no captions, no watermarks and no logos of any kind - it is a clean photo only.And its task prompt pins the output to the original photo:
Director LLM · task prompt
You receive a service brief and the BEFORE photo. Write ONE image generation prompt for the AFTER photo: the EXACT same scene, camera angle, framing and lighting direction as the BEFORE photo, all the same fixed objects and layout, but now showing the realistic result of the service described in the brief - cleaned, restored or transformed ONLY in the ways that service plausibly changes, nothing structurally added or invented beyond the service scope, same time of day, photorealistic, looking like it was shot on the same phone seconds later. Dramatic but believable.In our run, the Director produced a single paragraph describing "a photorealistic wide-angle interior photograph of the same child's bedroom after a professional deep clean," working through every element of the scene: the white iron bed frame now neatly made, the wardrobe closed, the shag rug vacuumed and clear of toys, the garland still hanging in the window, the same angle and warm natural light, and no text or watermarks anywhere.
The AFTER image generation node takes that generated prompt plus the before photo as a reference and runs it through an AI image generator to render the after visualization. The node lets you pick the model (Gemini 3 Pro Image, GPT-Image-2, or Zoviz Image Generate), the aspect ratio (1:1 up to 21:9 and vertical formats), and the resolution (1K, 2K, or 4K).
The SPLIT SHEET image node is driven by a fixed prompt that never needs editing:
Prompt · split screen image
One single split screen comparison image: the LEFT half is the FIRST reference photo (the before) exactly as it is, the RIGHT half is the SECOND reference photo (the after) exactly as it is, same scale and alignment, separated by a clean thin white vertical line, small clean text label BEFORE in the top left corner and AFTER in the top right corner, no other changes to either photo, vertical 9:16 composition.The VIDEO generation node connects to both the before photo and the after image, so the clip starts and ends on real frames. Its prompt:
Prompt · transformation video
A satisfying transformation reveal: start exactly on the first frame (the before state) and end exactly on the last frame (the after state). The scene transforms smoothly and continuously in between, dirt lifting away, surfaces brightening, order appearing, like a time lapse of the service being completed. Camera locked in place, no cuts, steady and seamless, satisfying calm pace. Soft ambient sound with a gentle whoosh as the transformation completes, no speech, no text.Want to generate a morph video like this one? Zoviz's AI video generator can turn a real before photo and a generated after image into one continuous transformation clip. Try our AI Video Generator here.
Want to generate a morph video like this one? Zoviz's AI video generator can turn a real before photo and a generated after image into one continuous transformation clip.
The Copy LLM node writes the posting kit. Its system prompt is short and strict: "You are a service business social media manager. Your entire reply is always exactly the labeled sections requested, nothing else. Keep every claim honest and modest." Its task asks for hook lines under 8 words, a caption under 150 characters ending with your CTA, one honest proof line for an on screen overlay, and 8 hashtags mixing the service niche, satisfying content, and local discovery tags.
How to run it yourself, step by step
The canvas ships with an instruction note pinned to the board, and the workflow follows it exactly:
- Drop your BEFORE photo into the IMAGE node. JPG or PNG only. The video step rejects webp.
- Fill in the SERVICE BRIEF with what your service really does: the service, your business name, what you did, where, the tone you want, the platform, and your CTA.
- Run the Director LLM, then the AFTER image node (about 7 credits together). Check the result and make sure it stays believable before you continue.
- Run the SPLIT SHEET image (6 credits), the Copy LLM (1 credit), and the VIDEO last (about 40 credits).
The golden rule from the canvas note: if you change the photo or the brief, rerun the Director LLM first. Every downstream node depends on its output, so running the image nodes against a stale prompt is the most common way to waste credits.
A full run costs roughly 54 credits, and about 40 of that is the video. A practical habit: iterate on the after image until you are happy, and only render the video once at the end.
Why the "same scene" constraint matters
Most AI image tools fail at before and after content for one reason: they redecorate. Ask a general image generator for "this room but clean" and you will get new furniture, a different rug, moved windows, a result any real customer would spot as fake in a second.
This canvas attacks that problem in three layers. The Director LLM is instructed to change nothing beyond what the described service would realistically change. The image generation node receives the original photo as a reference, not just a text prompt. And the video is forced to start and end on your actual frames, so the transformation reads as one continuous scene.
The pipeline also carries an explicit ethics note, and it is worth repeating: the after image is a visualization of your typical result. Only show what you can actually deliver. Used honestly, it is a preview of your service, not a fabricated portfolio.
Read More in our blog about How We Turned One Photo Into a Commercial Video With Zoviz Canvas.
Who this works for
We ran it as a cleaning business marketing example, but the brief is generic on purpose. Any service with a visible transformation can reuse the same canvas by changing one photo and one paragraph: home and office cleaning, car detailing, landscaping and lawn care, pressure washing, painting and decorating, home organization, furniture restoration, carpet and upholstery care, or junk removal.
Because the canvas is reusable, the second run takes about two minutes of setup: replace the photo, rewrite the brief, rerun the nodes in order.
Running a small service business? This canvas is one piece of a bigger toolkit. See 10 best AI tools for small businesses for other places AI can save time beyond marketing content.
FAQs
What is an AI before and after generator?
It is a workflow that takes a real "before" photo and generates the matching "after" image, comparison graphics, and video with AI instead of waiting for staged photography. This canvas is one example built from connected LLM, image, and video nodes.
Can I make a before and after video from just one photo?
Yes. The canvas first generates the after image from your before photo, then renders a morph video that starts on the real photo and ends on the generated after frame.
What is the best format for before and after ads?
Post the split screen image as the static ad or cover, and the morph video as the main creative. The 9:16 vertical format works across TikTok, Reels, and Shorts, and the pipeline generates both assets from the same pair of images.
Is it honest to use an AI generated after photo?
Treat it as a visualization of your typical result, not proof of a specific job. Keep the changes inside what your service really delivers, and say so in the caption when it matters. The canvas prompts are deliberately written to prevent exaggeration.
How much does a full run cost?
About 54 Zoviz credits: roughly 7 for the Director LLM plus after image, 6 for the split screen image, 1 for the copy, and about 40 for the video.
Try the canvas. Open the Before-After Transformation Ad canvas, clone it into your workspace, drop in one photo from your last job, and describe what you did. Five minutes later you will have an after visualization, a split screen before and after picture, a transformation video, and the caption to post them with. Clone the canvas →
