This is a simple, beginner-friendly workflow for creating fruit AI drama, dating shows, and cinematic short videos using ChatGPT and Pixero. You’ll go from a reference TikTok video to finished AI-generated scenes, then stitch them together into a final short.
Pick a fruit-style TikTok or Shorts video you want to recreate in your own version. You are not copying it literally. You are using it as a structure reference.
Save the video so you can study the cuts, scenes, expressions, and pacing.
The script controls the dialogue, emotional tone, and scene order. You want the exact lines or a cleaned-up version of them.
Usually every major cut becomes a new scene. Take a clean screenshot of each scene so ChatGPT can understand the visual structure.
Upload all the screenshots, paste your script, paste your scene breakdown, and then paste the master prompt from this guide.
Ask ChatGPT to analyze the reference, swap the fruit characters into new fruit identities, lock the cast, and generate image prompts for every scene.
Open Pixero using my link, go to AI Studio, choose your image model, set the aspect ratio to 9:16, and generate every scene image one by one.
Save them in order:
scene-01.jpg,
scene-02.jpg,
scene-03.jpg, and so on.
If a character looks wrong or loses consistency, use a reference image from an earlier good generation and regenerate that scene.
Once all your images are approved, go back to the same ChatGPT chat and ask it to generate the animation prompt for each scene.
Back in Pixero, use the scene image as your start frame, paste the animation prompt, choose the video model, and generate each scene video one by one.
Import all the final video scenes into CapCut or your editor, put them in the right order, add music if needed, adjust pacing, and export your final short.
Paste this into a new ChatGPT chat after you upload your screenshots and paste your script + scene breakdown.
I want you to help me recreate a fruit-style cinematic short video based on the reference material I upload. Your job is to act like a professional AI storyboard director and prompt writer. IMPORTANT GOAL: We are building a short-form fruit AI story for TikTok / Shorts. The workflow is: 1. analyze the reference material 2. identify the characters, story roles, and visual style 3. create NEW fruit characters inspired by the reference roles, but NOT using the exact same fruit types 4. lock the characters and world 5. generate image prompts for every scene first 6. after I approve the images, generate animation prompts scene by scene IMPORTANT: Do NOT copy the reference characters literally. Do NOT reuse the exact same fruit types from the reference unless I explicitly ask. FRUIT SWAP RULE: When analyzing the reference, preserve: - the character role - the gender presentation - the vibe - the relationship dynamic - the fashion energy - the emotional purpose in the story But change the fruit type into a new fruit choice. Examples: - banana husband in the reference can become lemon husband - strawberry wife in the reference can become peach wife - orange boss can become plum boss - grape child can become peach child The goal is: SAME STORY FUNCTION NEW FRUIT IDENTITY Do NOT make the new fruit swaps random. Pick fruit types that still match the energy of the character. STYLE RULES: - vertical 9:16 - premium cinematic 3D animated realism - fruit-headed characters - not human - not chibi - not toy-like - not doll-like - emotionally clear - high consistency across all scenes - stable identity, wardrobe, proportions, and world design - visually rich, cinematic, and TikTok-friendly - no text, no subtitles, no watermark, no logos unless I ask FIRST TASK: Analyze the uploaded reference material and the script I provide. Then identify: - the main story roles - the visual style - the world / location - wardrobe consistency - emotional tone - key props - camera language Then create: 1. a STORY STYLE SUMMARY 2. a ROLE-TO-FRUIT SWAP PLAN 3. a LOCKED CHARACTER SYSTEM 4. a LOCKED WORLD / LOCATION SYSTEM 5. SCENE-BY-SCENE IMAGE PROMPTS ONLY IMPORTANT CONSISTENCY RULE: Every scene prompt must repeat the full core character description for any character visible in that scene. Do not shorten the character descriptions too much. Consistency matters more than being brief. VERY IMPORTANT: If the reference has recurring characters, keep them visually locked across all scenes. If wardrobe should remain the same across connected scenes, keep it the same. If the reference uses close-ups and wides, keep the same character identity across all shots. OUTPUT FORMAT: Please respond in this exact structure: 1. STORY STYLE SUMMARY - short summary of the visual style, tone, and world 2. ROLE-TO-FRUIT SWAP PLAN For each main reference character, explain: - original role in the reference - new fruit replacement - why that fruit fits the role 3. LOCKED CHARACTER SYSTEM For each NEW character, write: - name - fruit type - gender / age feel - face details - fruit texture details - eye details - outfit - personality energy - continuity rules 4. LOCKED WORLD / LOCATION SYSTEM - describe the main locations and visual tone 5. IMAGE PROMPTS ONLY For each scene, provide: SCENE [NUMBER] — SOURCE IMAGE PROMPT [full image prompt] Use this structure inside every image prompt: - format and quality - location - visible cast only - scene description - camera - negative prompt if needed IMPORTANT: - Do NOT write animation prompts yet - Do NOT write voice timing yet - Do NOT write editing instructions yet - Do NOT summarize too much - Make every scene prompt copy-paste ready REFERENCE MATERIAL: [PASTE REFERENCE NOTES HERE] VIDEO SCRIPT: [PASTE FULL VIDEO SCRIPT HERE] SCENE BREAKDOWN: [PASTE SCENE-BY-SCENE BREAKDOWN HERE] EXTRA RULES: - preserve the story logic, but change the fruit identities - preserve the emotional roles, but do not literally copy the fruit characters - make the new cast feel original but clearly inspired by the reference structure - if the reference is melodrama, keep it melodramatic - if the reference is funny, keep it funny - if the reference is office-based, keep it office-based - if the reference is home-based, keep it home-based VERY IMPORTANT: First give me: - style summary - role-to-fruit swap plan - locked character system - locked world system - all image prompts scene by scene Then stop. Wait for me to ask for the animation prompts later.
Example:
Reference video style: - premium fruit-headed melodrama - vertical TikTok short - strong dialogue moments - recurring husband, wife, boss characters - cinematic lighting - emotionally exaggerated but visually polished
Paste the full cleaned-up dialogue or transcript.
Scene 1: Wife and husband in living room at night. Scene 2: Husband at office with boss. Scene 3: Husband comes home late. Scene 4: Boss reveals the truth.
After you approve all your images, paste this into the same ChatGPT chat.
Now use the locked characters, locked world, and approved scene image prompts we already created. Your task now is to generate the ANIMATION PROMPT for each scene. IMPORTANT: - keep the exact same characters, fruit types, wardrobe, world, props, and emotional continuity from the approved image prompts - do not redesign the characters - do not change the fruit types - do not change the scene order - keep the same visual style and consistency - dialogue is the most important part - make the dialogue sound natural, dramatic, and short-form friendly - keep each clip optimized for 8 seconds unless the scene clearly needs a different length - assume these will be used for AI image-to-video generation - make the prompts copy-paste ready For each scene, respond in this exact format: SCENE [NUMBER] — ANIMATION PROMPT [full animation prompt] And inside each animation prompt include: 1. clip length and format 2. identity preservation 3. audio lock 4. world 5. visible cast only 6. motion 7. speaker lock and dialogue timing 8. important voice rule 9. lip-sync rule 10. SFX / ambience only 11. end state 12. camera IMPORTANT WRITING RULES: - do not write weak generic dialogue - make the dialogue emotionally sharp and easy to understand - keep dialogue short enough to fit the clip naturally - if the scene is silent, say that clearly - if a scene should have no dialogue, do not force dialogue into it - keep the prompts clean and production-ready IMPORTANT: Do not regenerate the image prompts. Only generate the animation prompts for the approved scenes. Start from Scene 1 and continue in order.
Keep the pacing fast. If a clip feels too slow, trim the pauses. If the dialogue is the hook, let it land quickly.
In this video I show you how to recreate viral fruit AI dating shows step by step using ChatGPT and Pixero. We go from reference video, to script, to scene screenshots, to image prompts, to animation prompts, and finally to the finished short. If you want the exact prompts, comment fruit and I’ll send them.
If you want to put this HTML guide online for free later, the easiest options are:
index.html in a repo and publish itindex.html and publish it as a one-page static site.