Midjourney Omni Reference, Pika Swaps, F Lite, NotebookLM Mobile, Natasha Lyonne’s AI Film, Krea Topaz Upscaling, Suno v4.5, Qwen 3 and More
Creative Workflow Roundup: No fluff, no sponsors, no affiliate links, just this week's key AI + creative tech news and my unfiltered lab notes.
In This Week’s Roundup: Midjourney locks references, Pika swaps props without reshoots, and Freepik’s F Lite ends licensing headaches. NotebookLM jumps to mobile, Natasha Lyonne mixes generative sets with live action, and Krea teams with Topaz for billboard-ready stills while GPT Paint turns doodles into edits. Suno v4.5 sings richer, Alibaba’s Qwen 3 balances speed and depth, and fresh drops from Tavus, HeyGen, and ElevenLabs boost lip sync, voice mirroring, and real-time conversation.
Midjourney Omni Reference Ships
• Need to Know: Omni Reference lets you drag a sample image to lock character, object or palette across every prompt, eliminating tedious C-Ref tweaking.
• Lab Notes: Finally! I have been looking forward to this for awhile, I will be testing this.
Pika Swaps Rolls Out
• Need to Know: Pika users can now replace any object or face inside generated footage with a prompt, keeping lighting and motion intact.
• Lab Notes: Reshoot-free product variations in minutes.
F Lite Licensed Image Model
• Need to Know: Freepik and Fal.ai released F Lite, a ten-billion-parameter DiT trained on eighty million fully licensed images, available under Apache 2.0.
• Lab Notes: Safe-for-client style guides no longer need Getty.
NotebookLM Goes Mobile
• Need to Know: Google’s research assistant now has its own app on iOS and Android powered by Gemini 2.5 Flash, offering audio overviews in fifty languages.
• Lab Notes: Pocket summarization means fewer late-night doc dives.
Natasha Lyonne’s AI-Driven Film
• Need to Know: The Poker Face star will direct and act in Uncanny Valley, mixing generative worlds from Moonvalley with live action under her new studio Asteria.
• Lab Notes: Ethical AI branding moves from deck to set.
Krea + Topaz Upscaling
• Need to Know: Krea Enhancer now pipes images into Topaz Labs for sharpening and can upscale stills to twenty-two-thousand-pixel resolution.
• Lab Notes: Poster-size boards from tiny source files, now inside Krea.
Krea GPT Paint Debuts
• Need to Know: GPT Paint adds visual sketch notes and basic shapes that ChatGPT reads as edit instructions inside Krea Image.
• Lab Notes: Whiteboard-style comping without leaving browser.
Suno v4.5 Drops
• Need to Know: Pro and Premier tiers gain richer vocals, wider genre blends and smarter mash-ups—think midwest emo meets neosoul at one click.
• Lab Notes: Temp tracks now rival indie studio masters.
Alibaba’s Qwen 3 Family
• Need to Know: Open Apache-licensed models from 0.6B to 235B parameters merge fast chat mode and deep reasoning with dynamic budget control.
• Lab Notes: Very competitive space!
Hummingbird-0 Lip Sync Model
• Need to Know: Tavus’ zero shot model outputs ten-second lip-synced segments in about one minute, preserving identity across five-minute clips.
• Lab Notes: Drop-in dubbing for global ads.
HeyGen Voice Mirroring
• Need to Know: New Voice Mirroring lets your avatar mimic any tone and style for non-robotic speech generation.
• Lab Notes: Cast a single VO artist across endless characters.