Product demo and ads: prompt recipes

Product ads, app promos and showcases. The Silent UGC recipe is the one we have actually run through the product; the product-ad recipes are built from the repo template and the tool descriptions.

4 recipes on this page: 1 verified, 3 untested. Our videos are AI-generated and say so when we post them. Connect AutoShort first with the steps on the MCP page, then paste a brief into Claude, ChatGPT, Cursor or Gemini CLI.

How to read a recipe

  • Verified means a real video was made through the product from this recipe or its direct predecessor, and the card links the public video and states the honest result, including low view counts.
  • Untested means we have not rendered a video from it yet. Drafting, editing and pricing are free; rendering costs credits.
  • The brief builds a draft you read before anything is charged. The assistant should call autoshort_render_draft only after you say yes. Prices come from the live price list: see pricing and costs. You can try the script step free on the free script preview.

Honest limits for this channel type

  • Never claim income, results or customer reviews.
  • The video is AI-generated; say so when you post it.

Recipes (4)

A 30-second ad with a hook in the first three seconds and three real features. Swap in your own product.

Untested. We have not rendered a video from this brief yet. Rendering costs credits, so it has not been run; the draft steps (scenes, edits, price quote) are free.

Hook (nine words or fewer)

“One candle. Poured by hand. Cedar and vanilla.”

Frame 1, already in motion

A hand-poured candle burning on a wooden shelf, the flame moving, camera already easing in.

The brief to paste

Make a 30-second product ad for a small shop with AutoShort. Art style AUTOSHORTS, 9:16, English narration (language en).

Hook, my first spoken line and first caption: "One candle. Poured by hand. Cedar and vanilla."
Frame 1 is already moving: a soy candle burning on a wooden shelf, flame moving, camera easing in.

Scenes, one idea each:
1. Hook over the lit candle.
2. Feature 1: hand-poured in small batches. Close on wax cooling in a jar.
3. Feature 2: a cedar and vanilla scent. Close on cedar shavings and a vanilla pod.
4. Feature 3: a reusable glass jar. Close on the empty jar with a plant in it.
5. Call to action: "Find your scent at Ember and Oak".

Look of every image: photographic, warm candlelight on a wooden shelf, product close-ups, no people, no faces, no readable text or labels on the jar.
Ending: Clear call to action. Friendly, honest tone. No health claims, no fake discounts, no comparisons to other brands.
Ember and Oak is invented: replace the name and the three features with your own real ones.

Write the scenes yourself and show me the narration and the visual of every scene before anything renders. Keep the narration to about 70 words. Price it with autoshort_estimate_cost. Render only after I say yes.
Art style
AUTOSHORTS (Professional). Photographic product look. For your own real product with photos, use the product-ad-from-a-page recipe instead.
Length and format
30 seconds, 9:16.
Language
Narration: English (en).
Voice
Not selectable on the draft path: the product narrates with its default voice. Steer the delivery with short sentences and plain punctuation; premium ElevenLabs voices are a separate paid option (see the options table).
Caption look
Picked automatically through MCP. In the dashboard's Motion graphics (beta) style, the "Sunset" look is the first one listed for this genre.
Estimated cost
40 credits (about $0.63 on the Starter plan, $0.80 at the $0.02-per-credit list rate). Ask for the price first: autoshort_estimate_cost quotes it for free.
Tools it triggers
  • autoshort_list_options
  • autoshort_draft_video
  • autoshort_get_draft
  • autoshort_update_draft
  • autoshort_estimate_cost
  • autoshort_render_draft

Visual direction for the images

  • One close-up per feature. The product is the only subject.
  • Ask for a plain jar with no label. Generated labels come back as gibberish; add your real logo in your own edit.
  • Warm light in every scene to keep the six images consistent.
  • Keep the call to action in narration and captions, not in the picture.

Two variations

Problem first: “Your living room smells like nothing.”

Open on the problem, then the product as the fix; same three features.

Make a 30-second product ad for a small shop with AutoShort. Art style AUTOSHORTS, 9:16, English narration (language en).

Hook, my first spoken line and first caption: "Your living room smells like nothing."
Frame 1 is already moving: a bare shelf and a window with curtains moving, camera sliding along the shelf.

Scenes, one idea each:
1. Hook over the bare shelf.
2. A candle is set on the shelf and lit.
3. Hand-poured in small batches.
4. A cedar and vanilla scent.
5. A reusable glass jar.
6. Call to action: "Find your scent at Ember and Oak".

Look of every image: photographic, warm light, no people, no faces, no readable text on the jar.
Ending: Call to action. No health claims, no fake discounts.
Ember and Oak is invented: replace it with your own real shop and features.

Write the scenes yourself and show me the narration and the visual of every scene before anything renders. Keep the narration to about 70 words. Price it with autoshort_estimate_cost. Render only after I say yes.
15-second cut: “Hand-poured. Cedar and vanilla. Yours.”

Two features and the call to action in 15 seconds.

Make a 15-second product ad for a small shop with AutoShort. Art style AUTOSHORTS, 9:16, English narration (language en).

Hook, my first spoken line and first caption: "Hand-poured. Cedar and vanilla. Yours."
Frame 1 is already moving: a candle flame close-up, wax glowing, camera pulling back slowly.

Scenes, one idea each:
1. Hook over the flame.
2. Hand-poured in small batches; a reusable glass jar.
3. Call to action: "Find your scent at Ember and Oak".

Look of every image: photographic, candlelight, product close-ups, no people, no readable text.
Ending: Call to action only.
Ember and Oak is invented: replace it with your own.

Write the scenes yourself and show me the narration and the visual of every scene before anything renders. Keep the narration to about 33 words. Price it with autoshort_estimate_cost. Render only after I say yes.

Do not

  • Do not ask for a label, a logo or a price tag inside the image.

    What we saw: The long uppercase no-text block made the image model redraw garbage text on 3 of 11 trial videos; a short "Pure illustration with no writing in it." replaced it. (our note: howto.md, line 7.)

  • Do not invent features, discounts or statistics: narration adds figures even when told not to.

    What we saw: 6 of 13 prompts that said "don't invent statistics" still got a percentage figure in the narration. (our note: video-feedback-2026-10-17.md, line 76.)

  • Do not expect real product photos to be zoomed or cropped: ads built from your own product photos keep them letterboxed on purpose.

    What we saw: Product photos are letterboxed on purpose in product ads (a punch-in would crop 6 % per side of the customer's product). (our note: polish-all-lanes-2026-10-28.md, line 14.)

Also see Product demo and ads use case, all prompts and what each step costs.

The sentence on screen, then the finished result it made. The claim stays on what the tool does.

Untested. We have not rendered a video from this brief yet. Rendering costs credits, so it has not been run; the draft steps (scenes, edits, price quote) are free.

Hook (nine words or fewer)

“One sentence. This is what it made.”

Frame 1, already in motion

The sentence being typed in a plain text box, the cursor still moving.

The brief to paste

Make a 15-second one-sentence showcase with AutoShort. Art style AUTOSHORTS, 9:16, English narration (language en).

Hook, my first spoken line and first caption: "One sentence. This is what it made."
Frame 1 is already moving: the sentence being typed in a plain text box, cursor still moving.

Scenes, one idea each:
1. The sentence typed on screen.
2. The best two seconds of the result.
3. A second, different result from a one-word change.
4. The settings used: the length and the style, as plain words.
5. Closing frame: the sentence again, ready to copy.

Look of every image: clean dark interface look with large plain type, then photographic result imagery. For the typed sentence, the text is added by captions, not drawn in images.
Ending: Call to action as text only: "Try this sentence". No claim about views, followers or income.
Show only things the tool really did: no invented results.

Write the scenes yourself and show me the narration and the visual of every scene before anything renders. Keep the narration to about 33 words. Price it with autoshort_estimate_cost. Render only after I say yes.
Art style
AUTOSHORTS (Professional). The sentence is carried by narration and captions, not by generated text.
Length and format
15 seconds, 9:16.
Language
Narration: English (en).
Voice
Not selectable on the draft path: the product narrates with its default voice. Steer the delivery with short sentences and plain punctuation; premium ElevenLabs voices are a separate paid option (see the options table).
Caption look
Picked automatically through MCP. In the dashboard's Motion graphics (beta) style, the "Tech" look is the first one listed for this genre.
Estimated cost
40 credits (about $0.63 on the Starter plan, $0.80 at the $0.02-per-credit list rate). Ask for the price first: autoshort_estimate_cost quotes it for free.
Tools it triggers
  • autoshort_list_options
  • autoshort_draft_video
  • autoshort_get_draft
  • autoshort_update_draft
  • autoshort_estimate_cost
  • autoshort_render_draft

Visual direction for the images

  • Do not ask the image model to draw a text box with words; ask for "a blank glowing text field" and let captions carry the sentence.
  • Result scenes should be real outputs of a real run: generate the short first, then describe what it contains.
  • Label the result honestly as AI-generated.
  • No metrics, no earnings.

Two variations

Two results compared: “Same sentence. One word changed.”

Put the two outputs side by side in order, one after the other.

Make a 15-second one-sentence showcase with AutoShort. Art style AUTOSHORTS, 9:16, English narration (language en).

Hook, my first spoken line and first caption: "Same sentence. One word changed."
Frame 1 is already moving: a blank glowing text field, the cursor blinking, a word being swapped.

Scenes, one idea each:
1. The sentence and the swapped word.
2. Result A, two seconds.
3. Result B, two seconds.
4. Closing: "Try this sentence".

Look of every image: clean dark interface look, then photographic results, no generated text.
Ending: Text-only call to action.

Write the scenes yourself and show me the narration and the visual of every scene before anything renders. Keep the narration to about 33 words. Price it with autoshort_estimate_cost. Render only after I say yes.
Silent UGC version: “POV: you type one line and close your laptop.”

Use the Silent UGC lane instead; see the Silent UGC recipe.

Make a 15-second one-sentence showcase with AutoShort. Art style AUTOSHORTS, 9:16, English narration (language en).

Hook, my first spoken line and first caption: "POV: you type one line and close your laptop."
Frame 1 is already moving: a laptop lid closing over a glowing screen.

Scenes, one idea each:
1. Hook over the closing laptop.
2. The finished video appears on a phone screen, plain glow.
3. Closing text-only line.

Look of every image: clean dark room, a laptop and a phone, no faces, no text on screens.
Ending: Text-only call to action.

Write the scenes yourself and show me the narration and the visual of every scene before anything renders. Keep the narration to about 33 words. Price it with autoshort_estimate_cost. Render only after I say yes.

Do not

  • Do not draw a phone or laptop screen with real content: it comes back as gibberish. Use a plain glow and put the real screen in your own edit.

    What we saw: Silent UGC sample: the first scene's phone UI is AI-generated gibberish. (our note: quality-benchmark-2026-10-27.md, line 65.)

  • Do not promise views, followers or income in the narration or the caption.

    What we saw: No income or results claims, testimonial stats, percentages, "in 2 minutes", links, handles, hashtags or emoji are accepted in hooks and captions: the server refuses them. (our note: silent-ugc-2026-10-26.md, line 14.)

  • Do not open on a title card or a wide establishing frame.

    What we saw: Hook rule: first sentence at most 9 words, no scene-setting opener, a face or a concrete action in the first shot. (our note: hook-first-2026-10-26.md, line 22.)

Also see Product demo and ads use case, all prompts and what each step costs.

A fictional AI character reacts to a phone with a big hook, then your real screen recording plays. Silent, AI-generated, for your own app only.

Verified: a real video exists. Watch it on its public page.

Three videos were made through the product on 26 Oct (campaign cmvoxljcc0041jotp7c8a4wzf, faceless character) and all three passed our QA: 12.00 s, 1080x1920, no audio, hands normal, no gibberish text. 130 credits in total, no refund. We posted two to YouTube Shorts: "POV: you type one line and close your laptop" (shown here) got 35 views, 95.0 % swiped away; the Vietnamese-hook version (https://www.youtube.com/shorts/IHXGzrx3MkQ) got 66 views, 87.8 % swiped away, read on 28 Oct. The median of our channel then was 5 views. No signup or purchase came from either. Views include a few of our own.

Our note: silent-ugc-posts-2026-10-26.md, line 17-28,43.

Hook (nine words or fewer)

“POV: you type one line and close your laptop.”

Frame 1, already in motion

A faceless character's hands holding a phone, a thumb moving, the hook text already in the upper third.

The brief to paste

Make a Silent UGC promo for my own app with the autoshort_ugc tools.

App name: [YOUR APP NAME]. URL: [YOUR APP URL]. What it really does, in two plain sentences (at least 20 characters): [WHAT YOUR APP DOES].
Goal: awareness. Scene-2 caption, 60 characters at most: "[WHAT THE SCREEN RECORDING SHOWS]".
Character: faceless. A fictional adult, hands and over-the-shoulder only, grey knit sweater, silver watch. No face, no real or public person.
Screen recording: I will give you a public https link to an MP4 or MOV, 3 to 60 seconds, up to 50 MB. Upload it with autoshort_ugc_upload_recording.

Suggest hooks with autoshort_ugc_suggest_hooks, show me the list and let me pick. My favourite style is: "POV: you type one line and close your laptop" (use it only if it is true for my app).
Hook rules: 8 to 90 characters, no income or results claims, no statistics, no "in 2 minutes", no links or handles.
Preview exactly one variant (over-the-shoulder, lean-in), show me the preview image, and quote the price with autoshort_estimate_cost.
Render only after I say yes.
Art style
No art style: you choose a character style (faceless or a realistic fictional face) and a setting and gesture from fixed lists.
Length and format
12 seconds, 9:16.
Language
Narration: English (en).
Voice
None. The video has no audio track; add a sound in the platform app.
Caption look
Picked automatically through MCP.
Estimated cost
44 credits (about $0.70 on the Starter plan, $0.88 at the $0.02-per-credit list rate). Ask for the price first: autoshort_estimate_cost quotes it for free.
Tools it triggers
  • autoshort_ugc_upload_recording
  • autoshort_ugc_create_campaign
  • autoshort_ugc_suggest_hooks
  • autoshort_ugc_preview
  • autoshort_estimate_cost
  • autoshort_ugc_render_batch
  • autoshort_ugc_get_campaign

Visual direction for the images

  • Faceless is the safer choice: hands on a phone, over-the-shoulder, back view. The phone screen is shown blank or glowing, never legible.
  • Pick the setting and gesture from the lists the preview tool accepts (we used desk-pov with tapping, over-shoulder with lean-in, and kitchen-counter with pause).
  • Hook text sits in the upper third, white bold with an outline, up to four lines of 22 characters.
  • Scene 2 is your real screen recording on a dark caption band; make sure it shows the result, not a loading screen.

Two variations

Discovery hook: “I just found out you can type an idea.”

A first-person discovery hook instead of POV.

Make a Silent UGC promo for my own app with the autoshort_ugc tools.

App name: [YOUR APP NAME]. URL: [YOUR APP URL]. What it really does: [WHAT YOUR APP DOES, AT LEAST TWO PLAIN SENTENCES].
Goal: awareness. Character: faceless, hands only, desk point of view, tapping.
Hook (first frame): "I just found out you can type an idea." Change it if it is not literally true for my app.
Upload my screen recording first, preview one variant, quote the price with autoshort_estimate_cost, and render only after I say yes.
Another language, same clip: “Mình vừa biết chỉ cần gõ một ý tưởng.”

Same character and clip, different hook language; a re-caption of an existing variant is cheaper than a new clip.

Re-use my existing Silent UGC campaign [CAMPAIGN ID] with the autoshort_ugc tools.

Make one more variant with the same setting and gesture but a different hook, in Vietnamese: "Mình vừa biết chỉ cần gõ một ý tưởng".
Show me the preview, quote the price with autoshort_estimate_cost (a re-caption that reuses the clip is priced separately from a new clip), and render only after I say yes.

Do not

  • Do not ask for readable app UI in the generated scene: the first sample's phone screen was AI-made gibberish. The real screen is scene 2.

    What we saw: Silent UGC sample: the first scene's phone UI is AI-generated gibberish. (our note: quality-benchmark-2026-10-27.md, line 65.)

  • Do not choose a face character if you need reliably clean hands and mouths; hands are the main deformity risk and faces soften. Check the preview before you pay for the clip.

    What we saw: Faceless mode shows hands on a phone; hands are the main deformity risk and the phone screen is never legible. (our note: silent-ugc-2026-10-26.md, line 15.)

  • Do not put income, results, statistics, links, handles, hashtags or emoji in a hook or caption: the server refuses them.

    What we saw: No income or results claims, testimonial stats, percentages, "in 2 minutes", links, handles, hashtags or emoji are accepted in hooks and captions: the server refuses them. (our note: silent-ugc-2026-10-26.md, line 14.)

  • Do not post it as a native TikTok video: it is a promotional app ad.

    What we saw: It is a promotional app ad, the class of post TikTok did not test for our account: use it on Facebook, Instagram, YouTube and X. (our note: silent-ugc-posts-2026-10-26.md, line 9.)

  • Do not expect a free account to render it: a paid plan is required.

    What we saw: A FREE plan is refused at render with PAID_PLAN_REQUIRED; previews and hook suggestions need a positive credit balance. (our note: silent-ugc-2026-10-26.md, line 13.)

Related real videos (not made from this recipe)

  • Posted on X, 26 Oct (first hook version)

    Posted with the "Made with AI" label. X impressions on our posts that day were between 5 and 15 (median about 5); we did not record this post's own count, so there is no result to report.

    Our note: published.md, line 95.

Also see Product demo and ads use case, all prompts and what each step costs.

Give the tool a product page; it extracts details and images and writes the ad. You cannot dictate the hook, so check frame 1 and revise scene 1.

Untested. We have not rendered a video from this brief yet. Rendering costs credits, so it has not been run; the draft steps (scenes, edits, price quote) are free.

Hook (nine words or fewer)

“Open on the product already in use.”

Frame 1, already in motion

The product in use in the first second: a hand pouring, wearing, opening or using it, not a packshot on white.

The brief to paste

Make a 30-second product ad from this public product page with autoshort_generate_product_ad: [PRODUCT PAGE URL].

Language: en. Duration: 30. Art style: AUTOSHORTS.
First price it with autoshort_estimate_cost and tell me the minimum credits. Do not run it until I say yes.
When it is done, poll autoshort_get_video, then show me the finished video and the scene list.
Check frame 1 and the first spoken line: it must show the product in use and start with the benefit, in nine words or fewer. If it does not, use autoshort_revise_scene on scene 1 to rewrite the narration and the visual and show me the new draft before rendering again.
No health claims, no invented discounts, no comparisons to other brands. The video is AI-generated; do not say it is a real customer.
Art style
AUTOSHORTS (Professional). The product page photos are used as scenes, so the art style mostly affects generated fill-in scenes.
Length and format
30 seconds, 9:16.
Language
Narration: English (en).
Voice
Not selectable on the draft path: the product narrates with its default voice. Steer the delivery with short sentences and plain punctuation; premium ElevenLabs voices are a separate paid option (see the options table).
Caption look
Picked automatically through MCP.
Estimated cost
At least 40 credits (about $0.63 on the Starter plan, $0.80 at the $0.02-per-credit list rate). Ask for the price first: autoshort_estimate_cost quotes it for free.
Tools it triggers
  • autoshort_estimate_cost
  • autoshort_generate_product_ad
  • autoshort_get_video
  • autoshort_revise_scene
  • autoshort_render_draft

Visual direction for the images

  • The tool uses your product photos; they are letterboxed on purpose, not zoomed, so pick a page with clean photos.
  • Scene 1 is the only one worth rewriting: put the product in use there.
  • Ask for no text in generated scenes; your own photos carry any real packaging text.
  • Label the video as AI-generated when you post it.

Two variations

15-second cut: “Meet the one that finally works.”

Shortest allowed ad length for a single benefit.

Make a 15-second product ad from this public product page with autoshort_generate_product_ad: [PRODUCT PAGE URL].

Language: en. Duration: 15. Price it with autoshort_estimate_cost first and wait for my yes.
Then check frame 1: the product in use, and a first line of nine words or fewer. Revise scene 1 with autoshort_revise_scene if not.
Another language: “Now in your own language.”

Same page, output in a different supported language.

Make a 30-second product ad from this public product page with autoshort_generate_product_ad: [PRODUCT PAGE URL].

Language: [LANGUAGE CODE FROM autoshort_list_options]. Duration: 30. Price it with autoshort_estimate_cost first and wait for my yes.
Then check frame 1 and the first line as in the main recipe.

Do not

  • Do not expect punch-in zooms on your product photos: product ads keep them letterboxed so nothing is cropped.

    What we saw: Product photos are letterboxed on purpose in product ads (a punch-in would crop 6 % per side of the customer's product). (our note: polish-all-lanes-2026-10-28.md, line 14.)

  • Do not let the narration add claims or figures that are not on the page: it adds statistics unprompted.

    What we saw: 6 of 13 prompts that said "don't invent statistics" still got a percentage figure in the narration. (our note: video-feedback-2026-10-17.md, line 76.)

  • Do not accept an opening packshot on a blank background: a weak first frame is the usual reason people swipe away.

    What we saw: Hook rule: first sentence at most 9 words, no scene-setting opener, a face or a concrete action in the first shot. (our note: hook-first-2026-10-26.md, line 22.)

Also see Product demo and ads use case, all prompts and what each step costs.

More

Questions

How much does a Silent UGC promo cost?

The card computes it from the live price list: portrait, preview and one new-clip variant. Ask for the quote with autoshort_estimate_cost before you approve.

Can I use a real person in the promo?

No. Characters are fictional AI adults, either faceless (hands and back view) or a fictional realistic face, never a real or public person.