product-size
Skill0 starsChanged 18 days agoMITView on GitHub
Measure a real product (a can, a bottle, a jar, a phone) in millimetres in AI-generated stills and takes, by the eyes of the person holding it, and catch a product drawn too big, too small, stretched, or changing size between shots. Use before buying video from a still that shows a product, when someone says the product looks too big, or when planning a product shot.
In this repository4 other files it gives its agents.
- blind-reviewSkill
- recurring-castSkill
- scene-setSkill
- sound-mixSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 14
Write a product card, for example product.toml:
- Distinctive: line 23
Ask the user for the real size.
- Distinctive: line 33
FLAG: not in a hand (on a table, in the air).
- Distinctive: line 34
INFO: no face to scale it (an insert), or no product found.
- Distinctive: line 40
Start the shot with the product on a table at true size and let the take lift it.
- Distinctive: line 48
No face in the frame, no scale.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Product size
Image models draw products too big, and size words in a prompt ("a 12 oz can", "fits in her hand") don't hold it. On real reels a 66 mm can came out 1.25 to 1.6 times its size, and its size changed from shot to shot. A vision model asked "is this the right size?" mostly guesses. Measured, it's arithmetic: adult pupils sit about 63 mm apart, so the face of the person holding the product is a ruler.
Set up once
Write a product card, for example product.toml:
[product]
what = "a soda can" # the words the detector looks for
size_mm = [66, 122] # the real width and height
ref = "renders/can.png" # optional: the real product, for the review sheetAsk the user for the real size. Never guess it from a picture.
Measure every still before it becomes video
uv run scripts/size.py <still|clip> [...] --card product.toml --sheet size_sheet.jpg, inputs in cut order.
- PASS: within 0.8-1.2x of the real width. A product held at the chest sits nearer the lens than the eyes and reads up to ~15% big; the tolerance covers it.
- FAIL size: re-roll or fix the still. Never buy a take from it.
- FAIL jump: the product changed size by more than 15% from one shot to the next.
- FAIL shape: the product is stretched.
- FLAG: not in a hand (on a table, in the air). Its distance to the lens is unknown: judge it on the sheet.
- INFO: no face to scale it (an insert), or no product found. Judge it on the sheet or by eye.
The sheet puts the real product at true scale beside the one in the frame. Give it to the blind review: with the sheet, Claude judged size right 10 times out of 10; from bare frames, 5 out of 10.
When the product is too big
- Start the shot with the product on a table at true size and let the take lift it.
- Or paste the real product into the still at the measured scale before animating it. Hands over the product make this fragile: check the result at full size.
- A product close-up is a still of the real product with a slow push-in, never a generated take.
Prompt words rarely fix size. After two failed fixes, change the shot: hands out of frame, the product on the table.
Limits
- No face in the frame, no scale.
- The width stays the same at any tilt only for round products. A box turned edge-on reads narrow.
- Glass and very thin objects are found less reliably. Check the sheet.
- A face turned past ~60 degrees gives no scale.
- The ruler is an adult face: children and extreme close-ups read off.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

