One photo, six cinematic shots: Claude as your production assistant
Hey, it's Cindy 🌱 You upload one product photo, Claude analyses it like a DP would, and writes you six ready-to-paste camera prompts that turn that single still into six different moving shots.
btw the waitlist for AI System Club is now open! If you're keen on using AI to automate your content, video editing, job search, business or build AI agents, feel free to sign up (only if you want to hehe). Ps. the waitlist gets first dibs:)
💡 The honest division of labour, up front. Claude does not make the video. Anthropic's own documentation is blunt about this: Claude "is an image understanding model only. It can interpret and analyze images, but it cannot generate, produce, edit, manipulate, or create images" (Claude vision docs). What Claude is genuinely excellent at is the part that actually decides whether a shot looks cinematic: reading the light, the angle and the material in your photo, then writing camera direction a video model can follow. Claude directs. A video model films.
the shape of it
How the whole thing runs
📸 Upload one photo to Claude. Drag it straight into the chat window. Claude reads JPEG, PNG, GIF and WebP, up to 10 MB and 8000x8000 pixels, and you can put up to 20 images in a single message if you want it to compare options.
🧠 Run the production assistant prompt. This is the one below. It makes Claude analyse the image properly first (light direction, surface, hero angle, negative space) instead of jumping straight to a generic list of camera moves.
🎥 Get six ready-to-paste shot prompts. Each one is a complete image-to-video prompt written for a specific camera move, plus a note on what that shot is for, so you know which is your hook and which is your closer.
🌱 Render them, then save the list as a template. You paste each prompt into an image-to-video tool alongside the same photo. Then you ask Claude to strip the list back into a reusable template, which is the step that turns twenty minutes of work into a permanent asset.
Why the analysis step is not optional: a video model has to invent everything your photo does not show, and it invents better when the prompt already commits to where the light is coming from and what the surface is made of. Skipping straight to "dolly in on the bottle" is how you get the plastic-looking result people blame the model for.
copy this
The production assistant prompt
This is the exact one. Upload your photo first, then paste this as your message. Everything in brackets is yours to fill in.
🎬 the production assistant
You are my production assistant and director of photography. I have one still photo and no ability to reshoot it. Your job is to turn it into a six-shot cinematic sequence that a text-to-video model can render from this single image.
Start by analysing the photo properly and telling me what you see, in this order: the subject and its material (glass, matte plastic, brushed metal, fabric, skin), the direction and quality of the light and where the shadows fall, the camera angle and rough focal length the photo appears to have been taken at, the background and how much depth is actually visible, the colour palette in plain words, and the negative space you could move the camera into. Then name the single strongest feature of this product or subject, the thing a shot should be built around.
Now design six shots that work as a sequence, not six variations of the same move. Give each shot a purpose: one attention-grabbing opener, two that reveal detail or texture, one that shows scale or context, one that creates desire or mood, and one closing hero shot. Use real camera language and vary the movement across the six: choose from push in, dolly out, orbit or arc, trucking move, pedestal up or down, tilt reveal, rack focus, and slow handheld drift. For each shot tell me what it is for in one line, then give me the shot prompt itself as a copy-paste block.
Write every shot prompt so it stands alone with no reference to the other shots, in this structure: the camera movement stated first and precisely, then the subject and what it is made of, then the lighting matching the photo, then the environment, then the mood and pacing, then the technical finish (for example "shallow depth of field, 24fps, subtle film grain, photoreal"). Keep each one under 80 words, describe only what the camera does and what exists in frame, and never include cuts, dialogue, music or text overlays, because the model renders a single continuous take.
Finally, flag honestly which of the six are most likely to fail on this specific image and why, based on what you can actually see, for example fine text on packaging, reflective surfaces, hands, or a background too shallow to orbit into.
what you get back
What a finished shot prompt looks like
So you know whether Claude got it right, here is the shape of one good output. Notice that the movement is first and specific, the material is named, and the lighting matches the source photo rather than inventing a new studio.
🎥 example shot: the slow orbit
Camera slowly orbits 90 degrees around the subject from left to right at a fixed height, ending square to the front label. A frosted glass bottle with a matte ceramic cap rests on a pale travertine surface. Warm directional light from the upper left rakes across the glass, throwing a long soft shadow to the right, with a faint warm bounce filling the shadow side. Background is a clean out-of-focus cream wall. Calm, unhurried, premium. Shallow depth of field, 24fps, subtle film grain, photoreal.
If a shot comes back vague ("cinematic shot of the product, beautiful lighting"), it is because the analysis step got skipped. Reply with: "rewrite shot 3 using only what you observed in the photo analysis: name the material, the light direction and the surface." That one line fixes it almost every time.
the vocabulary
The camera moves worth knowing
You do not need film school, but you do need the right eight words, because these are the terms video models were trained on. Vague wording like "make it move nicely" gets you drift. These definitions follow standard film usage (StudioBinder's camera movements guide).
Push in The camera moves physically closer to the subject. Reads as growing emphasis or realisation. The most reliable single move for an opener.
Dolly out The camera pulls back to reveal context. Great as a closer, and it is forgiving, because the model has to invent detail rather than resolve it.
Orbit or arc The camera travels in a curve around the subject. The most impressive move from a still, and the most likely to warp, so use it on simple shapes.
Trucking The camera moves laterally, left or right, staying parallel to the subject rather than turning. Clean, calm, and good for showing a row or a set.
Pedestal The whole camera rises or lowers without tilting, so the height changes but the angle stays level. Quietly expensive-looking on products.
Tilt The camera stays put and pivots vertically. A tilt up a tall object is the cheapest reveal there is, and it rarely breaks.
Rack focus Focus shifts from one plane to another while the camera holds still. Needs genuine depth in the source photo, so check before you ask for it.
Handheld drift A slow, slightly imperfect float. Adds life and realism, and it hides model artefacts better than any locked-off move.
render them
Where to actually run the prompts
Claude writes the prompts; you paste them into an image-to-video tool along with the same photo. Here is the honest state of the options, because "free AI video" is the single most out-of-date claim on the internet right now.
🌱 Dreamina (dreamina.capcut.com) gives 225 free credits a day, which reset every 24 hours. Exports may carry a watermark depending on your export settings, resolution and model. It is the most generous free daily allowance of the three.
🌱 PixVerse (pixverse.ai) gives recurring free credits: check your in-app balance rather than trusting a number from a blog post, because the amount varies and is not published.
🌱 Vidu (vidu.com) gives a monthly free allowance, caps free output at 720p and adds a visible watermark. It is the strongest of the three at keeping one subject consistent across several clips.
💡 Kling (klingai.com) is excellent at camera control and is paid only. It no longer has standing free daily credits, so treat any tutorial that says otherwise as out of date.
⚠️ Read this before you use a single clip for a client. All three free tiers above are for personal, non-commercial use under their own terms. Free output is for testing and for your own feed, not for paid work or a client deliverable. If the video is going anywhere commercial, use a paid plan and check that plan's licence yourself. Free daily caps also mean a full six-shot sequence can take more than one sitting, or one sitting spread across all three tools.
the reusable part
Turn the list into a template you keep
This is the beat most people miss, and it is where the actual leverage is. The specific prompts are tied to this photo, but the structure of the six shots is not. Run this straight after and you never start from scratch again.
🌱 the reusable template
Now strip the six shots back into a reusable template I can apply to any future product photo.
Keep the six shot purposes and the six camera movements exactly as they are, since that sequence works. Replace everything that was specific to this particular photo with clearly marked placeholders: [SUBJECT AND MATERIAL], [LIGHT DIRECTION AND QUALITY], [SURFACE], [BACKGROUND], [MOOD]. Leave the technical finish line intact at the end of each prompt, because that part does not change between products.
Then write me a short filling-in guide: for each placeholder, tell me what to look for in a new photo and give me two or three example values, so I can fill it in accurately without re-running the whole analysis. Finish with a one-line instruction I can paste into a future chat alongside a new photo to have you fill the template in for me automatically.
Output the whole thing as a downloadable markdown file so I can keep it outside this conversation.
Save that file into a Claude Project called something like "product shots". Every future chat inside that project already knows your template and your brand's look, so a new photo becomes a two-minute job. Projects are on every plan, with up to five on a free account (Claude help centre), and Claude can build the markdown file for you on any plan too (file creation).
be realistic
Where this genuinely falls over
Every honest version of this workflow includes this section, because the failures are predictable and mostly avoidable once you know them. None of these are reasons not to try it. They are reasons to choose your photo well.
Text on packaging Labels, logos and small type warp or turn into nonsense the moment the camera moves. Choose shots where the label stays roughly front-on, or accept that close orbits will mangle it.
Highly reflective surfaces Chrome, mirrors and polished glass produce reflections that do not track correctly as the camera travels. Slow, small moves survive. Fast orbits do not.
Hands and faces If a hand holds your product, expect trouble on any move longer than a couple of seconds. Crop the hand out of the source photo if you can.
Flat backgrounds An orbit or rack focus needs somewhere to go. A product shot against a plain seamless wall has no depth to reveal, so the model invents it, usually badly.
Claude's own limits The docs note Claude's spatial reasoning and counting are approximate, and it will not identify people in images. Treat its read of the photo as a strong first draft to sanity-check, not gospel.
The sequence still needs you Six clips are not an edit. Ordering them, choosing durations and cutting on the movement is the part that makes it feel like a film, and that judgement is yours.