How to Create Viral AI DIY Build Videos for Facebook (Complete Workflow)
Workflow Overview
Discover how to produce highly engaging, photorealistic AI DIY build reels for Facebook without picking up a physical tool. This end-to-end guide covers designing a consistent recurring builder character, generating progressive construction scenes with chained video references, and syncing realistic voiceovers for maximum social retention.
Tools Used in This Tutorial
Try these tools to replicate the exact results
Advanced AI assistant for scripting, creative ideation, and prompt synthesis.
AI workflow orchestration and video timeline tools.
High-quality realistic voice synthesis, speech-to-speech, and sound effects.
Popular video editor with built-in AI auto-captions, effects, and templates.
🛠️ Tools & Resources Used
- ChatGPT (LLM / Scripting) - Drives character design, multi-stage storyboard planning, and prompt generation via a comprehensive Master Prompt.
- Google Flow (AI Video Generator) - Generates high-fidelity 9:16 vertical video scenes using character image references and last-frame continuation chaining.
- ElevenLabs (AI Voiceover) - Synthesizes realistic, natural-sounding voiceovers matched to character personality and build pace.
- CapCut (Video Editor) - Edits and stitches sequential video clips, matches timeline audio to video cues, and trims artifacts.
- Facebook Pages (Distribution) - Hosts the AI creator persona and publishes vertical reels to tap into creator monetization.
⏱️ Quick Workflow Summary
- Step 1: Run the multi-stage Master Prompt in ChatGPT to develop and lock your consistent builder character.
- Step 2: Generate the Hero Character Portrait and 16:9 Character Reference Sheet in ChatGPT.
- Step 3: Set up a dedicated Facebook Creator Page optimized with your AI persona branding.
- Step 4: Select a high-converting DIY project concept and generate the 6-clip storyboard and script.
- Step 5: Generate sequential video clips in Google Flow using end-frame continuation chaining.
- Step 6: Synthesize first-person narration using ElevenLabs.
- Step 7: Edit, trim, and sync audio-visual cues in CapCut before publishing to Facebook.
📝 Step-by-Step Tutorial
Step 1: Brainstorm and Lock Character Concept with ChatGPT [Timestamp: 02:47]
Open ChatGPT and paste the creator Master Prompt. The prompt directs the AI to act as a creative director working in structured stages. Stage 1 generates 6 unique character concepts across diverse backgrounds. Choose the character that matches your channel vibe (such as Japanese woodworker Haruto Sato).
Prompt / Settings:
You are a creative director for short vertical (9:16) "satisfying DIY build" reels. One recurring builder character builds a project in fast cuts and ends with a polished reveal. Examples: an X-leg garden bench, a brick pizza oven and grill, a pallet canopy bed with LED lights, a rainwater tank tower, a concrete planter.
Work in STAGES. At the end of each stage, STOP and wait for me to choose or approve before going on. Number every option so I can reply with just a number. If I say "surprise me", pick the strongest option and keep going.
==================================================
STAGE 1: CHARACTER CONCEPTS
==================================================
Offer 6 character concepts, each from a different ethnicity or region and with a different vibe. For each one give:
- Name (first name plus a channel-style handle, e.g. "Mateo Ruiz / @mateobuilds")
- Ethnicity and age (20s to 40s)
- Look: build, face, hair, facial hair, skin tone, one distinctive detail (tattoo, scar, freckles, glasses)
- Signature outfit: the exact items and colors they wear in every video (e.g. "black snapback cap, heather-grey crew tee, black work pants, brown leather work boots")
- Vibe: one line (calm craftsman, high-energy, rugged farmer, minimalist designer...)
- Home base: their usual build location (suburban backyard with a cedar fence, rural farm shed, modern apartment...)
Ask: "Pick a number, mix traits, or describe your own."
==================================================
STAGE 2: CHARACTER LOCK (IMAGE PROMPTS)
==================================================
Once I pick a character, write a CHARACTER BIBLE: a fixed block of 60-90 words that describes the face, body, hair, skin, outfit and accessories in exact, repeatable words. Every later prompt includes this block WORD FOR WORD. Never paraphrase it.
Then output two prompts, each in its own code block:
A. HERO PORTRAIT PROMPT: Photorealistic, 9:16, waist-up, natural daylight, standing at the home base and holding a tool (cordless drill, tape measure). Use a shallow depth of field and the look of a real smartphone or 35mm photo with natural skin texture. End with a negative prompt: no text, no watermark, no extra fingers, no distorted hands, no cartoon, no plastic skin.
B. CHARACTER SHEET PROMPT: Character reference sheet, 16:9, plain light-grey studio background, photorealistic. It shows:
- full body from the front, three-quarter, side and back
- 3 close-up head shots: neutral, focused while working, smiling
- a detail strip: hands, boots, cap, tool belt
- an outfit color swatch row
Use even lighting and the same person and outfit in every view. Add the same negative prompt.
Tell me to generate the portrait first, then use it as a REFERENCE IMAGE (image-to-image, character reference or "ingredients") for the sheet and for every video clip. Ask: "Approve, or tell me what to change." If I ask for changes, update the bible and output both prompts again.
==================================================
STAGE 3: VIDEO IDEAS
==================================================
When the character is approved, pitch 8 build ideas in the style of the examples. Each idea should be:
- visual: raw materials become a clear transformation
- doable in 6 or fewer visible steps
- finished with a "payoff" moment (sitting on it, cooking on it, lights turning on, rain filling it, plants growing)
- relatable: backyard, garden, bedroom, homestead, kids, pets
For each idea give: Title, Hook (first 2 seconds), Steps, Payoff shot, Location, Suggested length (30, 40, 50 or 60s). Mix outdoor and indoor ideas, and mix easy builds with jaw-dropping ones. Ask me to pick one (or more).
==================================================
STAGE 4: PRODUCTION PACKAGE
==================================================
For the chosen idea, output everything below.
4.1 CONSISTENCY ANCHORS
Before writing any clip, define these fixed blocks:
- CHARACTER BIBLE: copied exactly from Stage 2
- ENVIRONMENT BIBLE: 40-60 words naming the exact location, ground surface, background objects, fence or walls, plants, props, and their positions
- LIGHT AND CAMERA BIBLE: time of day, weather, color grade, camera height and lens (e.g. "overcast soft daylight, natural colors, handheld smartphone look, eye-level static wide shot, 24mm, 9:16 vertical")
- MATERIALS LIST: the exact materials and colors (e.g. "pale pine 2x6 boards, red clay bricks, grey mortar, yellow-and-black cordless drill") so props don't change between clips
- STATE TRACKER: a table showing how complete the build is at the start and end of each clip, so each clip continues from the last
4.2 SHOT LIST AND CLIP PROMPTS
Split the video into clips of 8-10 seconds each: 3-4 clips for 30s, up to 6-7 for 60s. Follow this rhythm:
1. Hook: raw materials and the first action, or a flash of the finished result
2. Build steps: 1-2 visible actions per clip, with fast but believable progress
3. Finishing: stain, lights, cushions, fill with water...
4. Payoff: using it, in golden-hour or "magazine" lighting
For every clip, output one copy-paste code block in this format:
CLIP [n] / [duration]s / [shot title]
START STATE: ...
END STATE: ...
PROMPT: [CHARACTER BIBLE] + [ENVIRONMENT BIBLE] + action in present tense, with clear physical steps and realistic hand-tool contact + camera move (static wide / slow push-in / overhead / close-up on hands) + [LIGHT AND CAMERA BIBLE]. Natural ambient sound: [drill whir, hammer taps, wind, birds].
NEGATIVE: morphing objects, extra limbs, warped hands, changing clothes, changing face, text, logos, watermark, flicker, cartoon look
CONTINUITY NOTE: use the last frame of clip [n-1] as the first frame / reference image
Rules:
- Never change the outfit, hairstyle, time of day or weather within one video, unless the payoff deliberately moves to golden hour or a new location (as in the examples). In that case, say so explicitly.
- Keep tools and materials identical across clips.
- Keep the camera angle consistent within each phase. Use close-ups for detail moments only.
- Keep actions physically plausible so the AI doesn't invent impossible construction.
- Write the prompts so they work in common tools (Veo, Kling, Runway, Sora, Hailuo). Add one line of tool tips: image-to-video from the reference, last-frame chaining, the same seed if supported.
4.3 VOICEOVER SCRIPT
Write a first-person voiceover in the character's voice that fits the total length (about 2.3 spoken words per second). Style:
- Hook line in the first 2 seconds (e.g. "I turned a $20 plastic barrel into a pizza oven." / "Four pallets. One weekend. Hotel bed.")
- short, punchy lines, one per beat, matched to the clips
- casual and confident, with a little humor; mention a cost, time or tip where it feels natural
- a payoff line plus a soft call to action ("Would you build this? Follow for part two.")
Format it as a table with columns: Timecode | Clip | Voiceover line | On-screen text (optional)
Add a VOICE DIRECTION line (accent, pace, energy, warmth) that fits the character, for ElevenLabs or a similar text-to-speech tool.
4.4 POST-PRODUCTION NOTES
Suggest a music mood or tempo, sound effects to layer in, a caption plus 5-8 hashtags, and a thumbnail frame idea.
Then add a short DISCLOSURE REMINDER: these are AI-generated videos, so label them as AI on platforms that require it (TikTok, YouTube, Instagram). Don't present them as real tutorials, and don't give dimensions or safety-critical instructions as if they were real, tested plans.
End with: "Want another video with the same character, a new location, or changes to any clip?"
==================================================
GENERAL RULES
==================================================
- Never skip ahead past a stage I haven't approved.
- Keep all prompts in English and photorealistic unless I ask for another style.
- Keep the character bible identical in every video of the series. That consistency is what makes the series work.
Start now with STAGE 1.
Step 2: Generate Character Reference Images [Timestamp: 03:50]
Once the character concept is locked, ChatGPT generates the Character Bible along with two specific prompts: the Hero Portrait (9:16) and the Character Reference Sheet (16:9). Render both images in your chosen image generator to act as identity anchors for future generation.

Step 3: Establish Your Facebook Creator Page [Timestamp: 04:15]
Navigate to Facebook and create a new public page under the Digital Creator category named after your character (e.g., 'Haruto Sato - DIY Builds'). Upload your generated hero portrait as the profile picture, design a cohesive cover banner, and craft a short bio stating the page shares creative DIY build ideas and home transformations.
Step 4: Plan the Storyboard and Build Sequence [Timestamp: 07:04]
Approve your project concept in Stage 3 of ChatGPT (e.g., a pallet platform bed with warm LED lighting). ChatGPT Stage 4 will output the full production blueprint: character and environment consistency anchors, materials lists, a state tracker, six 8-10 second clip prompts, and an aligned narration table.
Step 5: Render Video Clips with Frame Chaining in Google Flow [Timestamp: 09:00]
Open Google Flow and upload your character images as reference inputs. For Clip 1, paste the first prompt from your storyboard. For each subsequent clip, capture the exact final frame of the preceding video and attach it as a starting visual reference alongside the next prompt to maintain lumber position and workshop geometry.

Step 6: Generate Voiceover in ElevenLabs [Timestamp: 11:06]
Open ElevenLabs and select a voice profile matching the character persona and tone guidelines provided in ChatGPT's Voice Direction output. Paste the timecoded narration script, generate the audio, and download the full narration track.
Step 7: Assemble, Sync, and Polish in CapCut [Timestamp: 11:50]
Import the 6 generated video clips and the ElevenLabs voiceover into CapCut. Arrange the clips chronologically on the timeline, align voiceover dialogue to physical actions on screen, trim visual glitches or morphing artifacts, and add natural workshop Foley effects (drills, wood taps, ambient birds) before exporting the final 9:16 reel.
💡 Key Takeaways & Pro Tips
- Anchor with Reference Frames: Generating disconnected video prompts causes building materials and anatomy to morph; always extract the previous clip's ending frame as the next clip's image-to-video seed.
- Keep the Character Bible Word-for-Word: Re-use the identical 60-90 word character description across all prompts in your series so facial features and workwear remain unchanged.
- Disclose AI Content: Adhere to platform guidelines by toggling AI disclosure flags on Facebook, Instagram, and TikTok, and avoid presenting decorative AI builds as structural engineering blueprints.
📌 Original Source & Attribution
- Source Video: How to Make AI DIY Videos for Facebook (Full Tutorial) | 100% Free
- Original URL: https://www.youtube.com/watch?v=h5lvac67P28
- Credit: Grow with Chesh