Reference images keep faces consistent but ignore character size/scale — any way to control relative height?

Gemini API
, ,
1

Subject:

Reference images keep faces consistent but ignore character size/scale — any way to control relative height?

Post:

I’m rendering scenes with gemini-3.1-flash-image, composing multi-reference: character refs + location ref + prompt.

One character (Frankenstein’s Creature) should be ~1.4× the height of the man next to him. I cannot get any compose to respect that. Everything comes back ~1.1× — a tall guy, not a giant.

What I tried:

  • Prompting it — 15 phrasings: “eight feet tall”, “head reaches the roof beams”, “crown at his chest”, camera angles, poses. The model satisfies the words without scaling: asked for one huge hand enclosing another, it drew two normal hands. Asked for a framing crop, it just moved him closer to camera.
  • A reference image with the correct ratio. I generated a two-figure ref that actually measures 1.4× (verified, same floor plane) and fed it in. Faces, skin, clothes all transferred perfectly. Ratio came back 1.1×.
  • Same ref as the only input — 1.1×.
  • “Draw these two figures exactly as arranged in the reference” — 1.1×.
  • All of it again on gemini-3-pro-image — identical.

The weird part: a plain text-to-image prompt with no scene got the 1.4× ratio first try. So the model can draw it — it just re-derives sizes from scratch whenever it composes references into a scene. Identity survives, proportions don’t.

Asks:

  1. Some structured way to pin relative scale between subject refs — even something simple like subject B height = 0.6 of subject A, enforced like aspect_ratio is.
  2. Or a per-reference strength setting so a layout ref can be followed structurally.
  3. At minimum, a line in the docs. They promise consistent identity across up to 14 refs / 5 characters (true! it’s good!) — but nothing mentions that size and proportions aren’t part of what refs carry. Would have saved me a lot of testing.

Happy to share the repro pair — the 1.4× reference and the 1.1× compose from it, same subjects, same model.

2

Hi, TL;DR: Specific prompting techniques can help mitigate this issue largely.

We tried this and had similar observations. It seems like when you attach reference images, the model focuses on matching faces and clothing, but defaults both figures to normal human heights side-by-side.

Here is what works:

  1. Explicitly limit what the images are used for: Tell the model to use the reference images strictly for faces and clothing, and explicitly tell it not to use the reference images for body height or proportions.
  2. Use body alignment instead of numbers: Instead of saying “1.4x taller”, state where the shorter person’s head reaches on the giant (e.g., “the human’s head reaches only to the creature’s waist level”).

Example prompt structure:

  • “Subject A (from Image 1): Human scientist. Use Image 1 strictly for his facial features, glasses, and vest.”
  • “Subject B (from Image 2): Giant creature. Use Image 2 strictly for his facial scars, skin, and coat.”
  • “Important: Do not use the reference images for body height or proportions. Enforce an extreme size difference where the human’s head only reaches the creature’s waist level.”

Here is the exact prompt I tried which gave expected results:

A wide-angle cinematic shot in a Victorian laboratory of two characters standing side by side on the exact same level stone floor, framed in full body.

Characters:
- Subject A (from human_ref.png): A normal-sized human gentleman scientist. Use human_ref.png strictly for his facial features, spectacles, and tweed vest.
- Subject B (from giant_ref.png): A towering giant colossus. Use giant_ref.png strictly for his facial scars, pale skin tone, and ragged coat.

Spatial Layout & Physical Anchors (Strict):
- Extreme anatomical size disparity: Subject B is a massive giant whose torso alone is as tall as Subject A's entire body.
- Physical alignment: Subject A's head reaches only to Subject B's lower waist/hip level. 
- Background benchmark: Subject B's head stoops to avoid hitting the 9-foot wooden ceiling beams, while Subject A has 4 feet of open air above his head.
- Subject B looks down at a steep 60-degree downward angle at Subject A, who looks steeply upward.

Do not render them at equal heights or standard human proportions. Both standing upright on the same floor plane.

Next steps:

  • Try the prompt structure above with your character reference images.
  • Alternatively, generate the scene and 1.4x scale difference first with a plain text prompt, and then use a follow-up edit turn with the reference images to apply the faces and clothes.
  • You can find more details in our multi-turn image editing documentation.
  • If you still face issues, let us know.