In my video I said you can give Kling 4.0 your before photo and your after photo, tell it to start high over the old roof, push in slow and end on the new roof, and get an ad out of it. That's real. Kling 4.0 is an AI video app from Kuaishou that went global on October 16, and it takes a first frame, an optional last frame and up to 10 frames in between, for clips from 3 to 30 seconds.
Here's the part I didn't have time for in the video. The first time we made it, we ran one 30-second take through six frames, and after the first few seconds it drifted: the roof went soft and smeared, and that one take was 405 credits. What looked right was short shots, 3 to 5 seconds each, where the start frame and the end frame are the same picture with a small camera move between them, and the move is written into the prompt second by second. So that's what you're getting: the two-photo method, the exact prompts we used with blanks for your house, the settings, what it costs, and the mistakes that eat credits.
One rule before anything else: use your own real photos of a job you actually did. An ad shows your work. Kling gets to move the camera. It doesn't get to do the roof.
Start here: paste this into ChatGPT or Claude
Not technical? Start here. Copy this prompt into ChatGPT or Claude (the free versions work) and it walks you through the whole guide, one step at a time, in plain English. It asks before it assumes anything and checks with you before anything that costs money, reaches a real customer or deletes something. The rest of the page is the same guide, written out, if you'd rather do it yourself.
Just want the prompt from the video? It's under The shot prompts below.
Hi! Please be my patient guide for this step-by-step guide from Receipts Group: "Turn your before and after photos into a drone-style roof ad"
The guide is here: https://receiptsgroup.com/guides/kling-before-after-roof-ad (open it if you can; if you can't, everything you need is below).
About me: I run a roofing or home-service company and I'm not technical. Use plain words and short sentences.
My goal: turn my own real before photo and after photo of one job into a short vertical video ad with Kling 4.0 (an AI video app): a slow camera push in on the old roof that cuts to the new roof. Two or three short shots, not one long take.
How to help me:
1. One step at a time. For each step, tell me in a sentence what it's for and exactly what to do. Then stop and wait until I say "done" or tell you what I see.
2. Ask before you assume anything about my company, the house, the job or my photos. One question at a time.
3. If you use a technical word, explain it in plain English the first time (for example: prompt, first frame, last frame or end frame, keyframe, crop, aspect ratio, 9:16, pixels, 720p and 1080p, credits, drift, push in, match cut, layer, opacity).
4. Stop and check with me before anything that costs money (every video Kling makes spends credits; buying a membership or more credits), anything that reaches real customers (posting the ad, running it as a paid ad, putting a homeowner's house in an ad) and anything that deletes something (my original photos).
5. Use only what I tell you or show you. Never make up details about the house, the job or my company. If something's missing, ask me.
6. Real work only. Both photos have to be my own real photos of a job my company actually did. Never help me make an "after" picture with AI, or show a roof we didn't do. If I ask, remind me why: an ad has to show my real work. Kling only moves the camera.
7. If a Kling screen looks different from what you expect, don't guess at button names. Ask me what I see and work from that.
The steps, from the guide:
Step 1. Pick the photos.
I need a real before photo and a real after photo of the same house, taken from the same spot, facing the same way, at the same zoom, with the phone held the same way (upright is best). Ask me what I have. If my two photos weren't taken from the same spot, tell me the cut from old roof to new roof won't line up, and help me plan the next job instead: note exactly where I stand, shoot the before, then stand in the same place for the after. The uglier the before, the better. For the golden-hour look, the after should be shot late in the day when the sun is low. If I fly a drone on jobs, a high photo from the same spot before and after is the "start high" version. Remind me to ask the homeowner before their house goes in an ad, and to keep the house number and street signs out of the frame.
Step 2. Check that they line up (free).
Help me put the after photo on top of the before photo at half see-through (any photo or video editor that lets me change a layer's opacity). The roof edges, the ridge and the corners of the house should sit in the same place. If they're a little off, nudge or scale the after photo until they line up. Never paint or change the roof itself.
Step 3. Make the frames (free).
Crop to a vertical 9:16 shape, at least 1080 pixels wide, from the full-size photos (not screenshots). I need four frames:
- Frame A: the before photo, the whole house.
- Frame B: the same before photo cropped in closer on the middle of the roof, about 1.3 to 1.6 times closer, still 9:16.
- Frame C: the after photo with exactly the same crop as Frame B.
- Frame D: the after photo cropped in about 1.3 times closer than Frame C.
The easy way to get the same crop on B and C: keep both photos lined up as layers from Step 2, crop once, and export each layer. Help me name the files so I don't mix them up. Check that every frame is still at least 1080 pixels wide.
Step 4. Fill in the prompts.
Help me fill in the brackets in these two prompts with what's really in my photos (the house, the roof, what's on the left and right, the light). Describe only what the photos show. Keep the house words the same in both. Show me the filled-in prompts and wait for my OK.
Shot 1 (5 seconds, first frame A, last frame B):
Camera: [24mm drone lens / phone camera at eye level]. Starts [where you took the photo from, e.g. high in front of the house / across the street at eye level], looking at the front of the [one-story / two-story] [colour] [stucco / brick / siding] house with the [hip / gable] roof, the old roof in the centre of the frame: [what's wrong with it, e.g. worn grey shingles, dark streaks, a blue tarp on the upper left, missing shingles on the right], the [what's on the left, e.g. palm tree on the front lawn] on the left, the [what's on the right, e.g. garage] on the right. 0.0 to 1.0 seconds: the camera holds still at the start position. 1.0 to 3.0 seconds: a slow push in begins, moving straight toward the centre of the old roof, gently easing in, keeping the same heading and the same angle. 3.0 to 5.0 seconds: the push continues at a steady pace, the old roof growing to fill the middle of the frame, [the worst of it, e.g. the tarp and the missing shingles] coming into clear detail. Ends close on the old roof at the same angle, the roof filling the middle of the frame, matching the end frame. The house, the old roof, the lawn and the shadows stay fixed; only [the trees, e.g. the palm fronds] move gently in a light breeze. Photoreal, [the light in your photo, e.g. warm low sun from the left, long soft shadows], natural colours, crisp sharp detail, bright and clean. Smooth, stable, eased motion, level horizon. No text, no letters, no logos, no signs, no house numbers.
Shot 2 (3 seconds, first frame C, last frame D):
Camera: [the same lens as shot 1]. Starts close on the new roof of the [the same house words as shot 1], [the same angle as shot 1], the roof filling the middle of the frame: brand-new [colour] [architectural shingles / metal panels / tile] in crisp straight [courses / rows], the [what's at the edge, e.g. palm tree] at the [left] edge. 0.0 to 1.0 seconds: the camera holds still at the start position. 1.0 to 3.0 seconds: the slow push in continues straight toward the centre of the new roof, keeping the same heading and angle, then eases to a gentle stop, [the light, e.g. the warm low sun raking across the new shingles]. Ends closer on the new roof's front slope at the same angle, [e.g. the ridge near the top of the frame], matching the end frame. The house, the new roof and the shadows stay fixed; the [shingles] stay crisp and sharp; only [the trees] move gently. Photoreal, [the same light words as shot 1], natural colours, crisp sharp detail, bright and clean. Smooth, stable, eased motion, level horizon. No text, no letters, no logos, no signs, no house numbers.
Step 5. The settings in Kling.
On Kling's website or app, use image to video with the Kling 4.0 model. Shot 1: first frame A, last frame B, 5 seconds. Shot 2: first frame C, last frame D, 3 seconds. No keyframes in between. 1080p if I have a membership, otherwise 720p (4K isn't on for 4.0 yet). The video takes its shape from the first frame, so the 9:16 crop makes it vertical. One version at a time: each extra version costs the full amount again.
Step 6. The paper check, then the first test.
Before I spend anything, check each prompt with me: does it start where its first frame actually is, is there one main move with a pace word, are the steps timed and do they add up to the length, does it say what stays still, does it end on the end frame, does it describe anything that isn't in my photos, and does it ask for any words or logos? Then tell me Shot 2 is the cheapest test (3 seconds is about 40 credits at 1080p, at about 13.5 credits a second) and wait for my yes before I make it.
Step 7. Look at what came back.
Help me check it: it sits still for about a second first (normal, I'll trim it), the move is smooth, the house stays the same house, the roof edges stay sharp, and it ends on my end frame. If something's off, help me change one thing in the prompt and try again. It usually takes a couple of tries. Then do Shot 1 the same way (5 seconds is about 68 credits at 1080p), and wait for my yes again.
Step 8. Put it together.
In my video editor: trim the still second off the start of each shot, put Shot 1 then Shot 2 so the cut lands on the moment the roof turns new, mute the clips' own sound if they have any, and add my words, logo and phone number in the editor (never in the Kling prompt). Remind me not to call it drone footage if it isn't, and to turn on the platform's AI label when I post. Show me the plan and wait for my yes before I post anything.How the pieces connect
Two photos become four frames, the four frames become two short shots, and the cut between them does the before and after. Here's the whole thing on one screen.
- 1Your two photosA real before and a real after of one job, shot from the same spot→ two pictures that line up
- 2The framesCrop both to vertical, then crop in closer for each end frame→ four frames: A and B from the before, C and D from the after
- 3Shot 1 in KlingFirst frame A, last frame B, a slow push in on the old roof, 5 seconds→ the old roof growing in the frame
- 4Shot 2 in KlingFirst frame C, last frame D, the push keeps going on the new roof, 3 seconds→ the new roof, settling in
- 5Your editorTrim the still second off each, Shot 1 then Shot 2, your words on top→ the ad: on the cut, the roof is new
B and C are the same crop of two photos taken from the same spot. That's the whole trick: on the cut, the frame doesn't move and the roof is suddenly new.
Words you'll see
- Prompt: the instructions you type into Kling.
- First frame / last frame: the picture the video starts on and the picture it has to end on. Kling fills in the motion between them. Some screens call the last one the end frame.
- Keyframe: an extra picture somewhere in the middle that the video has to hit. Kling 4.0 takes up to 10. You won't need any for this.
- 9:16: the tall, vertical shape of Reels, TikTok and Shorts.
- 720p / 1080p: how sharp the video is. 1080p is sharper.
- Credits: what Kling charges per video, by the second.
- Drift: when the video slowly wanders off your pictures: the house changes shape, the roof smears, things appear that weren't there.
- Push in: the camera moves straight toward the subject.
- Match cut: a cut between two shots framed exactly the same, so only the thing that changed (the roof) jumps.
Step 1: two photos that line up
Stand in the same spot both times
Wherethe job, before tear-off and after cleanupPick a spot where the whole front of the roof shows, like across the street. Note exactly where you stood (the crack in the sidewalk, the mailbox), and stand there again for the after. Same direction, same zoom (leave it at 1x), phone held upright both times. Take a few each time, so you can pick the pair that matches best.
Shoot the before ugly
Don't tidy it up. The tarp, the streaks, the missing shingles: that's the story. The whole ad is the trip from ugly to brand new, so the uglier the before, the better it plays.
Shoot the after in good light
If you want the golden-hour look from my video, take the after late in the day when the sun is low. Asking Kling to change the light is one more change, and every change is a chance to drift. Shoot the light you want.
Drone photos? Same rule
If you already fly a drone on jobs, a high photo from the same spot before and after is the “start high over the old roof” version from the video. Same rule: same spot, same height, same direction. With ground photos, the shot starts where you stood, and that works fine.
Ask first, and keep the address out
I'd ask the homeowner before their house goes in an ad. Keep the house number, street signs and anything with a name on it out of the frame, or crop it out.
Step 2: make the four frames (free)
Check that they line up
Put the after photo on top of the before in any photo or video editor that lets you change a layer's opacity, and set the top one to about half see-through. The roof edges, the ridge and the corners of the house should sit in the same place. A little off? Nudge or scale the after photo until they match. Never paint or change the roof itself.
Crop to vertical
Crop to the 9:16 shape (tall, like a phone screen), at least 1080 pixels wide. Start from the full-size photos, not screenshots. Kling takes the video's shape from the first frame, so this crop is what makes the ad vertical.
Make the four frames
- Frame A: the before, the whole house.
- Frame B: the same before, cropped in on the middle of the roof, about 1.3 to 1.6 times closer. Still 9:16.
- Frame C: the after, with exactly the same crop as B.
- Frame D: the after, cropped in about 1.3 times closer than C.
The easy way to get the same crop on B and C: keep both photos lined up as layers from the last step, crop once, and export each layer. Check every frame is still at least 1080 pixels wide. A small or blurry frame gives you a blurry video.
Name them
Something like
A-before-wide,B-before-close,C-after-close,D-after-closer. Mixing up a first frame and a last frame means you pay for a video that runs backwards.
Step 3: the shot prompts
These are the prompts from the ad in my video, with blanks for your house. Fill in every [bracket] with what's really in your photos, and delete the examples. Use the same words for the house and the light in every shot: the model reads each prompt fresh, so the same words keep it the same house.
Camera: [24mm drone lens / phone camera at eye level]. Starts [where you took the photo from, e.g. high in front of the house / across the street at eye level], looking at the front of the [one-story / two-story] [colour] [stucco / brick / siding] house with the [hip / gable] roof, the old roof in the centre of the frame: [what's wrong with it, e.g. worn grey shingles, dark streaks, a blue tarp on the upper left, missing shingles on the right], the [what's on the left, e.g. palm tree on the front lawn] on the left, the [what's on the right, e.g. garage] on the right. 0.0 to 1.0 seconds: the camera holds still at the start position. 1.0 to 3.0 seconds: a slow push in begins, moving straight toward the centre of the old roof, gently easing in, keeping the same heading and the same angle. 3.0 to 5.0 seconds: the push continues at a steady pace, the old roof growing to fill the middle of the frame, [the worst of it, e.g. the tarp and the missing shingles] coming into clear detail. Ends close on the old roof at the same angle, the roof filling the middle of the frame, matching the end frame. The house, the old roof, the lawn and the shadows stay fixed; only [the trees, e.g. the palm fronds] move gently in a light breeze. Photoreal, [the light in your photo, e.g. warm low sun from the left, long soft shadows], natural colours, crisp sharp detail, bright and clean. Smooth, stable, eased motion, level horizon. No text, no letters, no logos, no signs, no house numbers.Camera: [the same lens as shot 1]. Starts close on the new roof of the [the same house words as shot 1], [the same angle as shot 1], the roof filling the middle of the frame: brand-new [colour] [architectural shingles / metal panels / tile] in crisp straight [courses / rows], the [what's at the edge, e.g. palm tree] at the [left] edge. 0.0 to 1.0 seconds: the camera holds still at the start position. 1.0 to 3.0 seconds: the slow push in continues straight toward the centre of the new roof, keeping the same heading and angle, then eases to a gentle stop, [the light, e.g. the warm low sun raking across the new shingles]. Ends closer on the new roof's front slope at the same angle, [e.g. the ridge near the top of the frame], matching the end frame. The house, the new roof and the shadows stay fixed; the [shingles] stay crisp and sharp; only [the trees] move gently. Photoreal, [the same light words as shot 1], natural colours, crisp sharp detail, bright and clean. Smooth, stable, eased motion, level horizon. No text, no letters, no logos, no signs, no house numbers.Camera: 50mm macro lens on a motion-control slider. Starts a few inches above the new [colour] [shingles] at a low grazing angle, looking up the roof slope, the [courses] running across the frame, [where the light comes from, e.g. the low sun in the top-left corner]. 0.0 to 1.0 seconds: the camera holds still at the start position. 1.0 to 4.0 seconds: a slow, steady slide to the right, parallel to the [courses], at the same height and angle, [e.g. the warm sidelight raking across the granules] and the crisp shadow line under each [shingle] edge, easing to a gentle stop. Ends on the same [courses] a little further to the right, matching the end frame. Focus holds on the nearest [course]; the [shingles] stay perfectly still. Photoreal, [the same light words as shot 1], natural colours, crisp sharp detail. Smooth, stable, eased motion, level horizon. No text, no letters, no logos, no signs, no house numbers.Writing your own
- Start with the camera, and start where your first frame actually is. If the photo was taken from the sidewalk, don't write “starts high.” The model has to invent the view, and it drifts.
- Time it. “0.0 to 1.0 seconds: holds still,” then steps of 2 or 3 seconds. Kling's own guide writes prompts in timed sections like this. The first second is a hold because Kling sits on the first frame for about half a second to a second and a half before it moves anyway, so you trim it.
- One main move per shot: a push in, a slide or a pan. Kling's camera guide says the same: stack a push, a pan and a tilt and it gets unstable.
- Put a pace word on every move: slow, steady, gentle.
- Say what stays still, or the model moves it.
- Say it stays smooth in plain positive words (“smooth, stable, eased motion”) rather than “no shaking.”
- End on the end frame: “matching the end frame.”
- Always finish with the no-text line. Video models warp letters, so your words go on in the editor.
The one-take version from the video (simpler, drifts more)
This is what I described in the video: one shot that starts on the old roof and ends on the new one. First frame A (the before), last frame C (the after, cropped in). Kling will let you set it anywhere from 3 to 30 seconds. Keep it at 5.
Camera: [24mm drone lens / phone camera at eye level]. Starts [where you took the photo from], looking at the front of the [one-story / two-story] [colour] [stucco / brick / siding] house with the [hip / gable] roof, the old roof in the centre of the frame: [what's wrong with it]. 0.0 to 1.0 seconds: the camera holds still at the start position. 1.0 to 3.0 seconds: a slow push in begins, moving straight toward the centre of the roof, gently easing in, keeping the same heading and the same angle. 3.0 to 4.0 seconds: as the push continues, the old roof is replaced by the brand-new [colour] [shingles] in one smooth sweep from [left to right], the walls, windows, trees and lawn staying exactly the same. 4.0 to 5.0 seconds: the push eases to a gentle stop close on the new roof, matching the end frame. Only the roof changes; the house, the yard and the shadows stay fixed. Photoreal, [the light in your after photo], natural colours, crisp sharp detail, bright and clean. Smooth, stable, eased motion, level horizon. No text, no letters, no logos, no signs, no house numbers.Step 4: the settings in Kling
| Setting | Pick | Why |
|---|---|---|
| Mode | Image to video, with a first frame and a last frame | The two frames pin where the shot starts and where it has to end |
| Model | Kling 4.0 | The version with first and last frames, keyframes and up to 30 seconds |
| Length | Shot 1: 5 s. Shot 2: 3 s. Shot 3: 4 s. One take: 5 s | Kling allows 3 to 30. Short stays sharp, and the first second is a hold you trim |
| Resolution | 1080p with a membership, otherwise 720p | 4K isn't turned on for 4.0 yet |
| Shape | Comes from your first frame | Crop to 9:16 first and the video comes out vertical |
| Keyframes | None | They're for big changes mid-shot. Our six-frame take is the one that drifted |
| Versions | One at a time | Each extra version costs the full price again |
| Sound | Mute it, if the clip has any | Your voice or music goes on in the editor |
You'll also see Omni-Ref, which mixes images, video, subjects and audio as references. You don't need it for this. And if you were hoping to wire this into your own app: Kling's API listed 4.0 as “coming soon” when I checked. The website and the app have it now.
First test: the paper check (free), then one 3-second shot
Before you spend a credit, check every prompt against its frames. Most bad takes are a prompt that describes something the frames don't show. You can do it yourself with the list in Step 3, or paste this into ChatGPT or Claude. Before you attach a photo: crop out the house number and anyone's face.
Please check this prompt for an AI video app (Kling 4.0) before I spend credits on it. The app animates a camera move between a first frame and a last frame, and both frames are my own real photos of a roofing job.
What my first frame shows:
[describe it, or attach it]
What my last frame shows:
[describe it, or attach it]
The length I picked: [3 / 4 / 5] seconds
My prompt:
[paste it here]
Check these, and give me a yes or no with one line on why for each:
1. Does the prompt start where my first frame actually is (the same spot, height and angle)?
2. Is there one main camera move, with a pace word like slow, steady or gentle?
3. Are the steps timed in seconds, starting with a 1-second hold, and do they add up to the length I picked?
4. Does it say what stays still?
5. Does it end on the end frame?
6. Does it describe anything that isn't in my photos? List it.
7. Does it ask for any words, numbers, logos or signs? It shouldn't.
Then give me the fixed prompt. Rules: change only what failed. Don't add anything that isn't in my photos, and never add to or change the work on the roof: it has to show the job exactly as it was done.Then make Shot 2 first. It's 3 seconds, the cheapest real test: about 40 credits at 1080p. Good: it sits still for about a second, moves smooth and slow, the house stays the same house, the shingle lines stay sharp, and it ends on frame D. Bad: the roof smears, windows move, something new shows up, or it never gets to D. Change one thing in the prompt and try again. Once Shot 2 is clean, make Shot 1.
Put it together and post
Trim the still second
Cut off the part where the camera sits still at the start of each shot. Let Shot 1 run all the way to its last frame, then start Shot 2 right before it starts moving. Same frame on both sides of the cut, new roof.
Put the cut on a beat
Land the roof change on a word or a beat in your music. In my video it lands on the word “new.”
Add your words in the editor
Your company name, your number, a BEFORE and AFTER label: all of it goes on in your editor, never in the Kling prompt. Mute the clips' own sound, if they have any, and put your voice or music on.
Label it honestly
Turn on the platform's AI label when you post. The roof is real and the camera move isn't, and people should know that. And don't caption it as drone footage if it isn't.
What it costs
| What | Kling credits at 1080p |
|---|---|
| Your photos, the frames, the prompts | 0 |
| One 3-second shot (Shot 2, the first test) | About 40.5 |
| One 4-second shot (Shot 3) | About 54 |
| One 5-second shot (Shot 1, or the one take) | About 67.5 |
| The two-shot ad, one try each | About 108 |
| One 30-second take (what we started with) | About 405 |
| Two versions of a shot at once | Double |
| A Kling membership or more credits, in dollars | Not checked for this guide: check Kling's pricing before you buy |
Mistakes to avoid
- One long take through a pile of frames. That was our 405-credit lesson. Short shots, each pinned at both ends.
- A prompt that starts somewhere your photo isn't. “Start high” with a photo from the curb makes the model invent the view.
- Stacking moves. Push in plus tilt plus orbit in 3 seconds is three chances to go wrong. One move per shot.
- Photos from two different spots. The cut jumps instead of matching.
- An AI after. If the roof in the ad isn't the roof you put on, it's a fake ad. Real photos only.
- Asking Kling for your logo or your number. It warps letters. The editor does words.
- Vague words: cinematic, epic, dynamic. Name the move instead: slow push in.
- Small or blurry frames. The video is only as sharp as what you give it.
- Re-doing everything when one shot is off. Fix the one shot.
When it breaks
The camera sits still for a second before it moves
The roof smears, warps or goes soft partway through
The house changes: a window moves, a tree shows up
It doesn't end on your last frame, or gets there early or late
The cut from old roof to new roof jumps
Letters or a fake logo show up on a truck or a sign
The video comes out sideways or the wrong shape
You can only pick 720p
The video looks blurry
The roof never turns new in the one-take version
A Kling screen doesn't match the guide
ChatGPT or Claude can't open the guide link
That's the prompt
Two real photos, two short shots, and the cut does the before and after for you. The uglier the before, the better it plays.
Checked on Oct 20, 2026 against
Every claim about a third-party tool in this guide (plans, prices, menu paths, commands, limits) was checked against these official pages on Oct 20, 2026. These screens change often: if something looks different, trust the page over this guide.
- Kling AI launches Kling 4.0 globally (press release, GlobeNewswire)Kling 4.0's global launch on October 16, 2026.
- Kling: introducing Kling 4.0Up to 30 seconds in one generation, up to 10 keyframes, Omni Reference (checked 2026-10-20).
- Kling: Kling 4.0 multiple keyframes guide3 to 30 seconds; start and end frames vs keyframes; prompts in timed sections; don't change everything in one keyframe; without milestones, events land too early or late (checked 2026-10-20).
- Kling: camera control guideOne main camera move per shot is usually better; stacked moves get unstable (checked 2026-10-20).
- Kling AIKling's website, where Kling 4.0 runs (checked 2026-10-20).
- Google Cloud: prompting guide for Veo 3.1The same first-frame and last-frame method on Google's video model: make both stills, then describe the move between them.
