Summary of "Nano Banana 2 + Kling 3.0 = Consistent Characters SOLVED"

Overview

This video shows how to improve AI character consistency in images and videos by using photorealistic character reference sheets together with an image model (Nano Banana 2) and a multi-reference video generator (Kling/Cling/Cling AI). The presenter demonstrates a step-by-step workflow, shares example prompts, and gives practical tips for creating consistent multi-angle images and more stable AI-driven video.

What the workflow achieves

Create hyperrealistic, multi-angle character images.
Reduce character drift in AI-generated videos by supplying explicit reference sheets.
Enable scripted camera cuts and multi-character interactions while preserving character identity.

Key concepts and product features

Character reference sheet (core idea)

A single reference image that contains 8 shots: 4 full-body shots (each facing a different direction) and 4 matching close-up face shots.

Layout: a 4-column vertical layout where each column corresponds to a camera angle (full-body + matching close-up).
Quality: the reference sheet should be high-quality and photorealistic; downstream images/videos inherit its realism.
Prompting tips: include camera and photography keywords to enforce realism (e.g., photorealistic, DSLR, 35mm, muted colors).

Nano Banana 2 (image generation)

Used to generate the character reference sheets and many consistent images of the same character(s) from different angles and distances.
Platforms mentioned: Higsfield AI (demo) and Google Gemini.
You can upload up to 14 reference sheets to combine multiple characters; consistency tends to degrade as more characters are combined.

Fixing common image-sheet problems

Common issue: both side-profile columns facing the same direction.
Fix: download the incorrect sheet, re-upload it to Nano Banana 2, and prompt the model to rotate or correct the problematic column (e.g., “change the full body shot in column 3 to face the opposite direction”).

Using your own characters

You can upload an existing AI-generated or real character image and prompt Nano Banana 2 to create a professional character reference sheet based on that upload.

Video generation (Kling/Cling/Cling AI and Omni reference)

Use a video tool that supports an “Omni reference” or multi-reference input to import character sheets and other reference frames.
Label each uploaded reference (for example: image1, image2, image3) and explicitly reference them in the prompt (e.g., “the woman from image3 and the man from image2 walk through image1”).
Script camera cuts, moves, and close-ups; the video model will leverage the reference sheets to reduce character drift.
Important prompt tip: explicitly state how the scene should end (for example, “scene ends with them in image1”) to avoid the generator cutting to the reference-sheet frame at the end.

Practical workflow — step-by-step

Write a character-sheet prompt that creates 4 body angles + 4 close-ups in a 4-column layout; include photorealism and camera keywords.
Generate the character reference sheet in Nano Banana 2 (Higsfield AI or Gemini).
Correct orientation/pose errors by re-uploading the sheet and prompting for specific fixes.
Upload reference sheets (and a starting frame image if needed) into the video generator’s Omni/multi-reference tool.
In the video prompt, map characters to specific reference images by label and describe movement, camera cuts, and the scene end to keep continuity.
Iterate — consistency is much improved but not perfect; minor face deformities can still appear.

Limitations and notes

Significant improvement in consistency, but not flawless: slight face proportion changes or imperfect matches may appear in some video frames.
More characters make perfect consistency harder to maintain.
Sometimes the video generator will cut to the reference sheet; prevent this by specifying the desired end-state in the prompt.
The presenter says the exact example prompt is available in the video description and references another tutorial with “14 practical Nano Banana tips.”

Guides, tutorials, and resources mentioned

Exact character-sheet prompt provided in the video description.
Another tutorial covering “14 practical tips” for Nano Banana (linked in the video).
Platforms: Higsfield AI (demo), Google Gemini (alternative).
Video generator: Kling/Cling/Cling AI (Omni reference feature).

Main speakers / tools referenced

Video presenter (unnamed) — walkthrough, example prompts, demonstrations.
Nano Banana 2 — primary image-generation model for character reference sheets.
Kling / Cling / Cling AI — video generation tool referenced for Omni/multi-reference support.
Higsfield AI — platform used in the demo for running Nano Banana 2.
Google Gemini — alternative platform mentioned.