Create AI images with multiple style references in Venice Studio

What style reference does

Style reference lets you feed Venice Studio images you like and generate brand new concepts that borrow their mood, color, and typography. You can pull inspiration from mood board sites like designinspiration.com or Behance, or use your own saved assets.

Start in the Venice Studio edit tab, where the style reference capability lives. Upload up to three images, then describe what you want built from them. A simple first pass is to blend loosely related images and see what synthesis comes back:

Create a mood board for this style I'm exploring

The model takes your prompt fairly literally, so treat the first result as a starting point and refine from there.

Merge references into a poster and clean up text

For a focused result, narrow to two references and give a concrete instruction. Here, one image supplies the subject and another supplies the typography:

A poster advertisement for this Yamaha motorcycle with typography like the first image. The text should say "Happiness is riding a motorcycle in the rain."

This example was run with Luma Uni 1 Max. Text often comes out with small glitches, so use the edit image toggle to fix just the broken part without disturbing the rest:

Remove the "UQ" text, but keep everything else exactly the same

The fix here used Nano Banana 2. Any model works, but more premium models generally give cleaner text and detail.

Tune per-image reference strength and iterate across models

Krea V2 large lets you set the strength of each reference image independently, so you can push some influences up and pull others down. If one image has the red tone you want and another has an orange you don't, raise the strength on the first and lower the third.

Create a photorealistic portrait of a Georgian woman in made-in-Georgia fabric, standing in the mountains outside of Tbilisi

Different models interpret the same prompt and strengths very differently. Use the "use prompt" icon to bring your prompt back, keep the same references, and re-run across models. Here, Krea kept returning the art style rather than photorealism, while Luma Uni 1 finally delivered the realistic portrait with the fabrics and detail intact. Iterating across models is part of the workflow, not a failure of it.

Upscale references with SeedVR2 before generating

Output quality tracks input quality: the higher the resolution of your reference images, the better the generated result. If a reference is pixelated, upscale it before using it.

In Venice Studio, choose the upscale option, select an upscaler model (SeedVR2 is the preferred choice here), and send it off. The upscaled image downloads to your computer automatically, ready to reload as a reference. Do this for every low-quality reference before you start generating so the style reference model has clean detail to work from.

Build world scenes with mixed references and strengths

For world building, load up to three cleaned-up references, pick a model, set an aspect ratio, and adjust each reference strength to favor the elements you like most. Venice currently offers four models for style reference; cycle through Krea V2 large, Krea V2 medium, Luma Uni 1 Max, and Luma Uni 1 to compare results.

Detailed prompts pay off here. An observatory scene, for example:

A grand circular observatory chamber with towering white marble columns. A mechanical model of the solar system with planets rendered as glowing geometric wireframe spheres with golden line patterns. No text.

When you want in-image text to match a reference, say so directly, for example "text in the sky matches image one typography." A powerful move is compounding: once you generate a scene you like, save it and use it as a reference for the next scene. Your browser stores generated assets, so you can select them straight from the asset picker and keep deriving new, connected scenes.

Turn your images into a world bible with agentic chat

Once you have a set of favorite images, take them into Venice agentic chat to build the story around them. Upload the images and ask the model to analyze and expand them into a coherent world:

Analyze these images and help me build a world. We want lore, mythology, characters, stories, etc., so we can make a world bible for my video game / film / novel / AI character.

The model reads the recurring visual language across your images (marble and gold filigree, floating wireframe shapes, water as sky) and proposes a foundation: cosmology, founding myth, central mystery, characters, and where it needs your input. This gives you a starting framework you can keep developing.

Beyond style reference and upscaling, Venice Studio can also combine images and do direct edits, like "change the tint of the image from blue to red and add a space nebula feel in the background." Because Venice does not log your prompt data and uses private models where possible, you can experiment freely.

Key takeaways

• Upload up to three references in the edit tab and describe the new concept you want built from their combined style.
• Use Krea V2 to control each reference's strength, and re-run the same prompt across models to find the look you want.
• Upscale pixelated references (SeedVR2) before generating, since higher-quality inputs produce higher-quality outputs.
• Compound your work by reusing generated scenes as references, then hand the set to agentic chat to build a full world bible.