Pinterest Affiliate Marketing Has Changed Forever with AI

You are an expert short-form video analyst, scriptwriter, and AI character designer, working inside a Claude environment with file and terminal access.
I'm giving you the file path to a downloaded reference video in my niche - it may run several minutes. I want you to:
1. Extract the audio and transcribe it (for example with ffmpeg to pull the audio track, then AssemblyAI to transcribe it)
2. Extract 6-10 evenly spaced frames from the video with ffmpeg, so you can see how it's shot, styled, and paced
3. Analyze the transcript and frames together to understand how the video is built - not to reuse its footage, script, or the real presenter's likeness in anything you produce
Do not identify, name, or describe the real presenter as a real, identifiable person - treat the frames only as a style reference for build, framing, energy, wardrobe, and setting.
YOUR TASK
Once you've extracted the transcript and frames, produce four things:
1. CONDENSED CONCEPT
The reference video may run several minutes. I only want a single 30-second video. Identify the strongest hook, single core demonstration, and closing call to action from the reference, and describe how to compress that into one continuous 30-second arc: a fast intro/hook in the first 3-4 seconds, one product demonstration through the middle, and a call to action in the final 3 seconds only. Do not try to preserve every scene from the original - pick only what earns its place in 30 seconds.
2. INITIAL CHARACTER PROMPT (for ChatGPT, GPT Image 2 Model)
One ready-to-paste prompt describing an ORIGINAL fictional presenter inspired by the reference frames - not a likeness of the real person. Change the specific facial features, hair, and exact wardrobe from what's shown, while keeping the overall vibe. The setting must match my niche (for example: a home hydroponic gardening setup, not a plain studio background), and must specify realistic skin texture, natural imperfections, and accurate anatomy. Request a single, front-facing, one-person image - not a multi-panel board.
3. FOLLOW-UP ANGLE PROMPTS (for the same ChatGPT conversation)
Four short follow-up prompts to send immediately after the first image generates, each requesting one additional angle of the exact same person, same outfit, same setting, as its own separate image:
- A 3/4 turned view
- A full side profile
- A rear view
- A close, front-facing portrait
Each follow-up prompt must explicitly say to keep the same face, hairstyle, outfit, and background as the previous image, and must ask for a single image, not a composite.
4. FEATURED PRODUCT
Name the specific product if it's identifiable, or the product category if it isn't, shown or referenced in the reference video, plus anything the transcript says about it. Don't invent a product that isn't actually shown or mentioned. I can also get the product myself, if you can.
OUTPUT FORMAT
Return the four sections above, in order, clearly labeled. Put the character prompt and each follow-up prompt inside their own code blocks so I can copy them directly.
NICHE: [e.g. hydroponic gardening]
REFERENCE VIDEO FILE PATH: [PASTE THE LOCAL FILE PATH TO YOUR DOWNLOADED VIDEO]

I'm attaching a real product photo from this item's Amazon listing. Clean it up into a single, plain product image:
- Remove all text, price badges, dimension callouts, arrows, icons, and  any other graphic overlays
- Keep the product itself completely unchanged - the exact same shape,  size, proportions, materials, colors, and every visible design detail
- Do not add, remove, or redesign any part of the product
- Place it on a plain, neutral background (white or very light grey),  evenly lit, no overlay elements
- Output one image containing only the product, nothing else

If I attach more than one photo from the listing, clean each one the same way and return them as separate images - do not merge multiple angles into one composite, and do not invent an angle that wasn't in one of my original photos.

Using the reference video's structure and hook style you identified above, and the character you already designed, write the Seedance 2.5 prompt for the actual video.

I found the product you flagged as featured in the reference video on Amazon, confirmed at [X] stars. Here's the exact listing:

PRODUCT TITLE: {Write the product name from amazon}
ABOUT THIS ITEM: {Paste product details from amazon}

Now create one complete, ready-to-paste Seedance 2.5  prompt for a single
vertical (9:16) video, generated in ONE Higgsfield call, following the same hook-demo-CTA arc you already identified. Don't re-explain or repeat your earlier analysis - just apply it.

HARD RULES
- Total runtime must not exceed 30 seconds.
- Structure the video as: a fast intro/hook in the first 3-4 seconds, one continuous product demonstration through the middle, and a clear call to action in the final 3 seconds only - do not spread the CTA earlier.
- Do not invent product specifications, results, prices, or claims beyond what's in the listing above.
- Refer to the presenter only as "@char_elena_v1" throughout the prompt - do not redescribe her face, hair, or wardrobe, since Higgsfield's Character tool already carries that identity. Only describe her actions, expressions, and positioning in each shot. Character’s images are attached for your reference.
- Describe the product reference precisely: preserve its exact shape, size, materials, colors, controls, and labels; the presenter must interact with it at a believable scale. If I'm supplying two cleaned product images (two real angles), treat them as the same object seen from different sides, not two different products. The two products images are attached.

VIDEO STRUCTURE
Build 5-7 clearly timed shots covering the full 30 seconds:
1. Hook (0-4s): a fast, attention-grabbing opening line and visual
2. Introduce and show the product
3. Set it up or prepare it for use
4. Demonstrate its main function
5. Show the result or benefit
6. Natural reaction or verdict
7. Call to action (final 3 seconds only): a short, direct line pointing at this pin's link/tag - never "linked below" or "in the description"

VOICE AND TIMING
Write the complete spoken script first, then give the exact word count and estimated duration at 125-140 words per minute, before the shot list. If it doesn't fit 30 seconds, shorten the script - don't ask for faster delivery to compensate. Include one brief natural pause.

SHOT-BY-SHOT DIRECTIONS
For every shot: start/end timestamp, shot purpose, camera framing, natural
camera movement, the character's action and expression, product action,
spoken dialogue, practical background sound, and the transition to the
next shot.

FORMAT AND REALISTIC UGC STYLE
- 9:16 vertical, mobile-first framing, presenter and product inside the center-safe zone (leave the top ~14% and bottom ~20% clear for Pinterest UI and captions)
- Realistic home environment appropriate to the product and niche
- Soft natural lighting, slight handheld movement, natural pauses and gestures
- Believable materials, reflections, shadows, physics, and consistent lighting across shots

AVOID
Redescribing the presenter's appearance, robotic camera movement, exaggerated expressions, salesy narration, invented product claims, generated captions baked into the scene, CTA content appearing before the final 3 seconds, and anything implying a result, rating, or price beyond the listing above.

OUTPUT FORMAT
First, outside the code block: the full spoken script with word count and duration check. Then, in one code block: the complete ready-to-paste Seedance prompt with the 5-7 shot timeline, camera direction, dialogue, and negative constraints.