claude skills / AI Video Ads / video-ad-prompts / en français

Video ad prompts.

AI-generated ad shots that look filmed, and a product that stays identical from one shot to the next.

See what it produces
  • Veo
  • Kling
  • Sora
  • Seedance
  • Instagram Reels
  • TikTok
40base prompts sorted by use case, ready to adapt
10elements per prompt, always in the same order
1action and one camera move per shot

orders of magnitude, depending on your volume

the problem

What gets in the way today.

  • 01The product changes shape, color or label from one generation to the next.
  • 02Hands come out with six fingers and the logo turns into unreadable letters.
  • 03You burn credits on vague prompts that chain three actions at once.

input, output

You give this. You get this.

What you give Claude: the product, real photos, the audience and the single message

Product: hand-thrown stoneware mug, 350 ml, 6 mm walls (product sheet)
Photos: mug-sand-front.jpg, mug-sand-texture.jpg, mug-in-hand.jpg
Audience: people who drink their morning coffee slowly
Use: product ad, 9:16, 15 seconds
Models: Kling for product shots, Veo for atmosphere
Look: warm cream and sand tones, soft window light, calm and tactile

$ claude "Write the shot list and the video prompts for a 15-second 9:16 ad for this mug, adapted to Kling and Veo."

Prompt set, excerptexample

Video prompts: Fennick Clay Studio, stoneware mug, product ad, 9:16, 15 s

Message: For people who drink their morning coffee slowly, the Fennick mug keeps a coffee warm in hand longer because its walls are thick, hand-thrown stoneware.

STYLE BLOCK (pasted into every prompt)

Soft morning window light from the left, warm cream and sand tones, gentle film grain, shallow depth of field, calm and tactile, no text, no letters, no logos, no watermark.

SHOT LIST

1 Hook, 2 s, text-to-video, Veo

2 Product, 3 s, image-to-video from mug-sand-front.jpg, Kling

3 Proof, 3 s, image-to-video from mug-sand-texture.jpg, Kling

4 Use, 4 s, image-to-video from mug-in-hand.jpg, Kling

5 End plate, 3 s, image-to-video, Kling

SHOT 2: PRODUCT

Master prompt:

The mug stands still on a light oak table. The camera slowly pushes in over five seconds toward the rim. Soft morning window light from the left, a thin highlight along the rim. Warm cream and sand tones, gentle film grain, shallow depth of field. No text, no logos. The mug's shape, glaze and speckles stay identical to the reference. 9:16, 5 seconds.

Kling version:

Slow push-in on the mug. Static mug, only the camera moves. Soft window light from the left. Keep the mug identical to the image. No text. 9:16, 5 s.

Takes: 4 | Pass: handle shape unchanged, speckles stable, no warping at the rim.

ADDED IN THE EDIT

Caption shot 3: "6 mm hand-thrown walls" (source: product sheet)

End card: real logo file, "Shop the mug", held 1.5 s

GENERATION LOG

Shot 4, take 4: fail, extra finger in 3 of 4 takes > filmed for real with a phone

the method

What Claude does, in order.

  1. step 1Set the message and the shots

    One sentence, "for whom, what benefit, thanks to which real fact", then 4 to 7 shots with their role and final length.

  2. step 2Pick the route for each shot

    Text-to-video for atmosphere, image-to-video from a real photo for the product, or a real shoot when that is safer.

  3. step 3Write with the anatomy

    Frame, subject, action, setting, camera, light, style, sound, exclusions and format, 40 to 120 words, then a version adapted to each model.

  4. step 4Plan the selection

    Number of takes per shot, pass criteria and a generation log, then the list of what the editor adds.

demo in French / click, try it

Your turn.

An interactive demo of the skill. It is in French, the pack works in any language: ask Claude to answer in English.

Product ad

the film / 44 seconds, with sound

The pack film.

the pack film

In 8 seconds.

the guardrails

What the skill never does.

  • The model never draws the logo or on-screen text: they are added in the edit from the real files.
  • The product starts from a real photo whenever the model allows it, otherwise the shot is marked approximate.
  • Nothing invented: no fake reviews, fake results or fake prices, and no real faces or brands without the rights.
  • Every shot makes sense without sound, and each model's current limits must be checked before promising a result.

Questions

Does the skill generate the videos itself?

No. It writes the prompts, the generation plan and the selection criteria. You run the generations in the tool of your choice, then the editor puts it together.

Why not let the AI write the text and the logo?

Because video models write badly and distort logos. The skill keeps them out of the prompts and places them in the edit, from the real files.

Are the prompts adapted to each model?

Yes: one master prompt, then one version per target model, longer or more literal depending on what that model follows best.

What about a UGC-style video or a testimonial?

The library covers those uses too. For a testimonial, the AI produces the scene, but the words come from a real customer with their consent, or the clip is presented as a reenactment.

the pack

In the pack: AI Video Ads.