The 6-Phase Workflow
Follow these phases in order. Each step builds on the last. The secret is in Phase 3—the storyboard method that keeps your character consistent across every shot.
The Structured Prompt
Don't skip this—it's the foundation of consistency.
- Upload your reference photo (e.g., you at a piano bar) to ChatGPT
- Use this exact command: "Turn this into a Nano Banana Pro prompt with sections: I. Subject, II. Setting & Context, III. Action & Mood, IV. Technical & Style. Add STRICT CONSISTENCY REQUIREMENTS."
- Clean the output: remove duplicates, simplify actions, ensure face/clothing rules are explicit
STRICT CONSISTENCY REQUIREMENTS:
- Maintain exact facial structure of reference image
- Do not alter clothing color or style
- Keep lighting temperature consistent (warm/moody)
- Mouth closed unless specifically directed
- No drastic pose changes between frames
MOTION RULES:
- Hands remain natural and relaxed
- No sudden head rotations exceeding 30°
- Maintain eye contact direction
One clean "Master Prompt" that locks in your face, clothing, and lighting for every subsequent generation.
The "Anchor" Image
One perfect still becomes your reality anchor.
- Open Higgsfield AI and select Nano Banana Pro as your model
- Upload your real photo as the reference image (drag and drop)
- Paste your Master Prompt from Phase 1 into the prompt box
- Generate a small batch (3-4 images) to give yourself options
- Pick the Hero Image: correct face, correct lighting, right mood
Your hero/base image. This is your "master look" for character, lighting, and environment. Every subsequent shot references this.
The Visual Storyboard
⭐ The Secret Sauce — Turn one good image into an entire cinematic sequence.
The Problem: Generating shots separately = different face every time.
The Solution: Build everything from a single hero image and a single storyboard.
- Start from your hero image: Take the best hero frame you generated in Higgsfield (Nano Banana Pro). This is your "master look" for character, lighting, and environment.
- Send it to Gemini for story planning: Paste the hero image into Gemini. Ask Gemini to imagine a shot-by-shot sequence and lay it out as a storyboard grid:
Tell Gemini you want a cohesive storyboard with 6-9 panels in one image, all based on that same hero look.
Shot 1: Wide shot — Empty bar, subject standing on stage Shot 2: Medium — Walking toward microphone, moody shadows Shot 3: Close-up — Eyes lifting to meet the mic Shot 4: Detail shot — Trembling hands on the mic stand Shot 5: Medium — Sitting at piano, fingers hovering Shot 6: Wide — Final pose, spotlight isolation - Choose your storyboard: Pick the storyboard grid that best matches your narrative and camera ideas. Download that single storyboard image (the multi-panel grid).
- Bring the storyboard back into Higgsfield: Upload the storyboard grid into Higgsfield. Use the Nano Banana Pro model and prompt:
"Extract each panel in this storyboard into individual 4K images. Keep facial details consistent. Use Panel 1 as the main face reference." - Get your final frames: Higgsfield now gives you separate, clean stills with different camera angles, same character, same room and lighting. These frames are ready for animation.
6-9 perfectly related shots, all born from one hero image and one storyboard. This is why the final video feels cinematic instead of random.
Animation (The Kling Pass)
This is where your stills start breathing.
You already have perfect frames from Higgsfield. Kling's job is ONLY to move the camera and the body subtly—not to reinvent the character.
Golden Rule: Do NOT re-describe who they are or what they look like. Only describe how the shot moves and what tiny actions happen.
"Man singing at piano in a bar, wearing a suit, dramatic lighting"
"Slow push-in toward the subject. Subtle micro-movement of eyes looking up toward the microphone. Hands tremble slightly on the mic stand. Cinematic shallow depth of field, no big body movement."
Recommended Settings in Kling:
- Duration: 5-10 seconds per shot
- Turn OFF "enhance prompt" if your motion prompt is already detailed
- Keep identity/style locked by relying on the input frame, not the text, for how they look
- You can turn audio on or off depending on whether you want ambient sound in the clip
Best Micro-Acting Prompts:
- A slow lift of the eyes
- A tiny swallow or deep breath
- Fingers tightening on the mic
- A barely-there turn of the head
- A small shift of weight from one foot to the other
A set of 5-10 second video clips, one for each storyboard shot. Raw footage ready for assembly.
The Editor's Eye
AI generates 5 seconds. Usually only 2 are usable.
The "Cherry Picking" Process:
- Trim the Glitch: AI videos morph weirdly at start/end. Cut those frames ruthlessly.
- The "Take" Selection: If you generated 3 versions of a hand movement, stack them in your timeline, compare, keep the most human one.
- Pacing: Don't let clips linger. Quick cuts hide imperfections and maintain energy.
Eyeline Match: Ensure directional continuity (if subject walks right in wide shot, they're still oriented right in close-up).
A trimmed, paced timeline with only the best seconds from each generation.
Polish & Export
- Arrange clips in narrative order: Establishing → Action → Emotion → Resolution
- Add music/sound design
- Export
A coherent, cinematic scene built from one reference photo and a clear workflow.
