How to recreate the Mr. Kumar reel — minimal shoot, maximum impact
Tools you'll need
- Camera / phone
- ChatGPT (Image Gen)
- Flow by Google
- CapCut
- Premiere Pro
Two types of clips
Type 01 — Talking / Dialogue Clips
You record yourself speaking the lines. AI transplants you into the cinematic dark executive office — full relight, new background, same you.
Type 02 — Pose / Still Clips
No talking — just a pose. ChatGPT recreates the reference pose with your face and body. Flow then brings it to life with cinematic motion.
01 · Dialogue Clips — Talking to Camera
You record the lines. AI puts you in the cinematic office.
Step 1 — Record your dialogue clips
Record each line of dialogue as a separate clip. Any background is fine — any lighting, any room. The AI will replace everything later. No studio, no green screen needed.
One sentence per clip. Short clips are easier to sync in editing later.
Step 2 — Take a screenshot from your clip
Open any clip you've already recorded. Pause on a clear frame showing your face and upper body. Screenshot that frame. You already have this from shooting the dialogue clips — no extra shoot needed.
Step 3 — Upload to Flow (Google) + paste prompt
Go to Flow by Google. Upload two things: your original dialogue clip + the background reference image. Select the model and paste the prompt below.
I am providing you with TWO inputs:
My real video clip — a person speaking directly to camera
A reference image — a cinematic dark executive office with a US map on the wall
SUBJECT: Extract and preserve the exact real person from the video. Keep their real face, skin tone, facial hair, expressions, lip movements, and speech intact across every frame. Do NOT generate a new person.
BACKGROUND: Remove the video background entirely. Replace it with the exact environment from the reference image — the US map, dark office atmosphere, desk lamp, leather chair, dark wood.
FRAMING: Zoom out to match the reference image — show full upper body, both arms, generous environment visible around the subject.
HANDS: Hands fully visible, fingers pressed together in a natural steepled/joined gesture.
LIGHTING: Dark cinematic relight — strong directional key light on face from above, deep shadows on body sides, warm lamp glow on left shoulder only. Reduce overall exposure significantly.
COLOR GRADE: Deep blacks, warm skin tones, teal/dark blue shadows, slightly desaturated environment, high contrast.
OUTPUT: The real person speaking naturally, in the exact reference environment, consistent and flicker-free across every frame.Reference Background
Use a cinematic dark executive office image (US map on wall, desk lamp, leather chair, dark wood) as your reference. Upload it to Flow alongside your clip for every dialogue generation.
02 · Pose Clips — Stills Brought to Life
No talking. ChatGPT recreates the pose. Flow animates it.
Step 1 — Screenshot the reference pose
Find the exact pose you want to recreate from the Mr. Kumar video. Pause it and screenshot that frame. This is your pose reference.
Step 2 — Screenshot yourself — from a clip you already shot
No need to shoot anything new. Go back to your dialogue clips from Phase 01 — find a frame where you're in a clear pose and screenshot it.
You already have this footage. Reuse it. Any frame with good posture works.
Step 3 — ChatGPT writes the prompt first
Upload both screenshots to ChatGPT (reference + yours). Ask it to write a generation prompt — don't ask it to generate yet.
I have uploaded a reference image and a photo of myself. I want to recreate the reference pose with the same lighting, same pose, same cinematic aura — but with my body, my face, and my clothing. It should look cinematic and luxury. Write me a detailed image generation prompt to recreate this. Do NOT generate the image yet — just write the prompt.Step 4 — Open a new chat — generate the image
Copy the prompt ChatGPT wrote. Open a brand new chat. Paste the prompt + upload both images again. Generate.
New chat = fresh context = better results. Always generate in a clean chat.
Step 5 — Upload to Flow — animate the still
Take the ChatGPT-generated pose image to Flow by Google. Upload the image only (no video this time). Paste the animation prompt below.
Animate the image while preserving the exact original frame, identity, pose, composition, and lighting. Do not alter the lighting in any way.
Camera Movement: Very slow cinematic pan left (first half), smooth transition to slow pan right (second half). Extremely subtle — as if shot on a professional motorized slider. Optional 1-2% push-in over the full duration.
Subject Motion: Natural micro-movements only — subtle breathing in chest/shoulders, occasional natural blink, tiny head stabilization, very slight finger micro-adjustments, minimal body sway (almost imperceptible).
Lighting: PRESERVE the exact original lighting. Do not relight. No moving shadows. No exposure changes. No flickering.
Style: Ultra-realistic. Premium cinematic commercial quality. Luxury entrepreneur portrait. 24fps cinematic motion blur. Duration: 5-8 seconds.
NEGATIVE: No lighting changes, no flickering, no face morphing, no pose changes, no camera shake, no warping, no AI artifacts, no identity drift.Phase 03 — Edit & Assemble
You have all the clips. Bring them into CapCut or Premiere Pro and follow the reference video structure.
- Import all dialogue + pose clips
- Watch the Mr. Kumar video — match clip order & timing
- Trim and sync — each dialogue clip on its spoken line
- Add the same background music as the reference video
- Colour grade — dark, cinematic, matching the office
- Export and post to Instagram Reels
Full workflow at a glance
Type 01 — Dialogue Clips
Record clip → Screenshot from clip → Flow: clip + bg image → Cinematic dialogue clip
Type 02 — Pose Clips
Screenshot reference + yourself → ChatGPT writes prompt → New chat: generate image → Flow: animate image → Animated pose clip
Made by @chloesinyin. Follow for more Ai content.