New tools for ultimate character consistency
Plus simple & effective video prompt tips
Reading time about 4 minutes.
Welcome to Visually AI!
I’m back from Greece as production wraps up for House of David, season 2! And I’m on to the next adventure, including post-production very soon.
I’ll be attending Freepik’s Upscale Conference in San Francisco on May 20-21, 2025.
Use this code to get 30% off your ticket: HEATHER30
🔮AI News this Week
We have some new tools for character, objects, and scene consistency with Runway and Midjourney.
Runway References
Runway’s new References image generation lets you upload a single reference image to maintain consistent characters across scenes in multiple poses.
You can prompt to continue in the same setting, focus on a specific area of the background, or match the color and tone.
Here are some of my first results using the same character in different settings animated by Gen-4:
Runway References has unique capabilities that allow you to be more flexible with your output, beyond consistency. You can expand the scene!
I used a frame from one of the videos generated above to get a different view: “Zoom in to the island in the background”
So now I have a new scene in the same world to use as an insert or transition: “Crane down and push in to the Greek island as the sun reflects off the deep blue waters’ surface”
You can continue working with your project however you choose. Here’s a longer video I made with the Greek coastal theme (it is not perfect, but I wanted to expand the original setting as an example):
Midjourney Omni-Reference
Midjourney launched the new Omni-Reference (--or) parameter for version 7 and it works on much more than character consistency.
You can use it on the web app or in Discord and adjust the reference weight from 0 - 1000 (100 is default).
Simply drag an image into the prompt bar "Omni-Reference" section on the website and adjust the slider to your preferred strength. On discord, type --cref and the image url and use --ow to change the strength.
Higher Omni-Reference weights increase adherence to your reference image, and you can use the parameter with personalization (--p), moodboards, style references (--sref), and stylization.
So far, I'm impressed with the consistent results. Here is my first set of results with the same character:
You can also use Omni-Reference to place objects in the background, on subjects, or transfer the style, color, or mood. I put the dream catcher image on the subject’s t-shirt in the example below:
This is a quick video I made using Omni-Reference, with the subject and style of the first clip as the reference image. I used the reference weight set to —ow 150 to keep the retro glam style from the first image:
📸 AI Snapshots
Google launched Gemini 2.5 Pro (I/O edition) with improved intelligence and coding ability, to use in Canvas.
LTXV by LTX Studio: The latest open source video model with camera motion control, multi keyframe conditioning, and multiscale rendering for consistent frame-to-frame coherence.
Leonardo AI is back in the image to video game with new camera motion controls with features such as, Smooth Video, Negative Prompt, Vibe, Lighting, and Color Theme.
Wan Video is now free and unlimited for all users in Relax mode & free Priority generation with credit, including daily credits for membership plans. The plans are currently 50% off: Pro $60/year, Premium $240/year (yes, that says per year).
Verified university students can get one year of free Cursor Pro, the AI coding tool.
🚀 My Recent Top AI Tool Picks
Dia: New open-source text-to-speech model with realistic, expressive voices.
Smithery: Find and install MCPs for AI agentic tasks from a collection of more than 5,000 options.
Your.mov: Upload uncut video content and describe the type of video you want. Get a fully-edited video with b-roll, captions, and other features.
Shumer Prompts: Browse through a collection of powerful LLM prompts - described as ‘GitHub for prompts’.
Higgsfield AI: Dropped even more incredible VFX effects and camera motions to help you create mind-blowing videos.
🖼️ Image Prompts
Prompt: Compact electric car parked in a clean, modern suburban neighborhood, cozy houses in the background, soft morning light, family-friendly feel, vibrant colors, realistic details
Prompt: Watercolor painting of a Japanese street with traditional houses and lanterns at night, neon lights reflecting on wet pavement, with a large moon in the sky.
🎥 Video Prompts
I’ve been experimenting with simple, direct prompts for image to video. Take a look at your image input and decide what you want to see:
Prompt: She blinks and turns her head towards the window, stands up and walks out of view
Prompt: She smiles confidently, tilts her head back and laughs knowingly while maintaining her regal posture
Prompt: She has a bored expression as she waits for her Uber to arrive. Cars drive by on the street behind her and a soft breeze blows her hair and shirt collar gently
Prompt: Dolly zoom out as the waves crash and flow onto the mystical checkerboard, stormy skies, realistic water physics, muted colors
Sound added with MMAudio on Replicate
Thank you for reading.
Have a creative week!
Looking good from Greece!