Creative Workflow Lab

Creative Workflow Lab

Share this post

Creative Workflow Lab
Creative Workflow Lab
Veo 3 with Audio, Google Flow Filmmaking, Gemini 2.5 Pro Boost, Imagen 4 Sharpens Up, Coding Agents, and More Creative AI Updates

Veo 3 with Audio, Google Flow Filmmaking, Gemini 2.5 Pro Boost, Imagen 4 Sharpens Up, Coding Agents, and More Creative AI Updates

Creative Workflow Roundup: No fluff, no sponsors, no affiliate links, just this week's key AI + creative tech news and my unfiltered lab notes.

Shannon Leonard's avatar
Shannon Leonard
May 25, 2025
∙ Paid
Share

In This Week’s Roundup: Google’s Veo 3 is very high-quality and adds groundbreaking native audio but still faces limitations like short clip lengths and some visual glitches. Flow emerges as Google’s timeline-based filmmaker tool, and Imagen 4 finally solves typography. New coding agent Jules promises hands-off code commits, while Gemini 2.5 Pro’s Deep Think mode tackles hallucinations head-on. FLUX models hit Azure, and Runway CEO dismisses Veo 3 as “hobby grade,” touting a superior Gen-4 workflow. NotebookLM brings AI research to your phone, Fal becomes Veo 3’s first external API host, and SynthID launches an essential detector portal. Plus OpenAI’s eye-catching acquisition of Jony Ive’s hardware startup io, Cartwheel’s Pixar-powered animation tools, Meshy’s instant rigging, and a must-save camera-move cheat-sheet.

Veo 3 Adds Native Audio
• Need to Know: Google DeepMind’s Veo 3 generates high-quality clips with synced voices and effects—announced at I/O 25.
• Lab Notes: Silent‑film era officially over; expect TikTok floods.

Creators Push Veo 3 Limits
• Need to Know: Artists share experiments from dance loops to anime; common gripe is eight‑second cap.
• Lab Notes: Flow’s storyboard chaining may be the workaround.

Google Flow Filmmaking Tool
• Need to Know: Flow offers timeline‑style prompting, shipping first to Pro and Ultra tiers.
• Lab Notes: Need to test this out ASAP.

Veo 3 “Still Not Perfect”
• Need to Know: @fofrAI thread lists artifacts: jelly limbs, clipping, exposure flicker.
• Lab Notes: Good checklist for client expectation setting. But let’s be clear, Veo 3 is a big step forward.

Ultra Plan Price Math
• Need to Know: Veo 3 sits behind Google AI Ultra at $250 / mo for 12 000 credits—about $3.12 per eight‑second 720p clip.
• Lab Notes: Budget accordingly!

Kling Lip‑Sync on Replicate
• Need to Know: Replicate added Kling v2’s lip‑sync model.
• Lab Notes: Replicate keeps adding useful models.

Meshy Auto‑Rigs Characters
• Need to Know: Meshy now auto rigs and animates 3‑D meshes in seconds, demoed with smooth walk cycles.
• Lab Notes: Opens indie path to mocap‑quality previs.

Stop “Vibe Coding” Chaos
• Need to Know: Vasu Man Moza’s thread lays out a four‑step system to replace ad‑hoc “vibe coding”.
• Lab Notes: Worth scanning before your next Claude refactor binge.

Overlap Video Agent Launches
• Need to Know: Overlap AI unveils an autonomous agent that cuts long videos into short viral clips and posts them for you.
• Lab Notes: Early users report hour‑long exports trimmed in under five minutes.

Jules Asynchronous Coding Agent
• Need to Know: Google quietly opened the Jules beta. Gemini 2.5 Pro plans fixes (and more) and submits PRs.
• Lab Notes: Looks powerful, but interested in how this compares to Codex.

NotebookLM Mobile App Rolls Out
• Need to Know: Google’s AI notebook hits iOS and Android; syncs sources and generates podcast‑style audio overviews.
• Lab Notes: Great for on‑set research or instant briefs.

FLUX Models Come to Azure
• Need to Know: Microsoft adds Black Forest Labs’ FLUX image models to Azure AI Foundry for faster, cheaper image generation.
• Lab Notes: A welcome escape hatch when GPU demand spikes.

Imagen 4 Nails Typography
• Need to Know: The Verge confirms the new Google Imagen 4 renders crisp text and detailed fabrics; rolling out in Gemini, Vertex, and Slides.
• Lab Notes: Finally safe to mock up posters without manual edits.

Fal Hosts Veo 3 First
• Need to Know: Fal says it is first external API for Veo 3 plus MusicGen and Imagen 4.
• Lab Notes: Pay‑per‑credit beats subscription for occasional use.

Google MusicGen Model Launch
• Need to Know: DeepMind releases Lyria RealTime for interactive composition; available in Vertex and Fal APIs.
• Lab Notes: Scratch track generator for trailers.

Director Workflow Showcase
• Need to Know: Google posts Flow short film made for Tribeca, highlighting storyboard‑to‑final pipeline.
• Lab Notes: Proof that AI-integrated media with humans-in-the-loop can hit festival quality fast.

Gemini 2.5 Pro Upgrade
• Need to Know: New preview scores top on GPQA and AIME.
• Lab Notes: Stronger math means more reliable coding and more.

Gemini Diffusion for Text
• Need to Know: DeepMind introduces a diffusion‑style language model claiming smoother long‑form generation.
• Lab Notes: Could cut hallucination rates.

Agent Mode in Gemini App
• Need to Know: Verge spotlights upcoming Gemini “Agent Mode” that can execute tasks end‑to‑end.
• Lab Notes: Mobile assistant may finally schedule shoots for you.

Deep Think Reasoning Mode
• Need to Know: New “Deep Think” toggle in Gemini 2.5 Pro parallelizes chain‑of‑thought for tougher prompts.
• Lab Notes: Early tests cut hallucinations on budgeting spreadsheets.

Runway CEO Throws Shade
• Need to Know: Cristóbal Valenzuela calls Veo 3 “still hobby grade” compared to Gen‑4 workflow.
• Lab Notes: Competitive bar rising fast.

Cool Runway Gen‑4 Workflow
• Need to Know: Runway CEO’s thread outlines reference‑image fine‑tune pipeline.
• Lab Notes: The reference age is here.

SynthID Detector Web Portal
• Need to Know: DeepMind launches free SynthID Detector site; drag‑and‑drop to spot AI‑generated regions in images or video.
• Lab Notes: Handy before shipping client deliverables.

Mistral Releases Devstral
• Need to Know: Devstral open model tuned for coding agents lands on Hugging Face; claims 70B quality at 8B size.
• Lab Notes: Could power on‑prem Git flows.

Cartwheel Exits Beta
• Need to Know: Fast Company reports Cartwheel’s AI animation tool raises $10 M seed and hires Pixar vets.
• Lab Notes: Promises hundred‑fold speed‑ups for 3‑D spot work.

AutoCaption Model on Replicate
• Need to Know: Makes karaoke‑style subtitles from any MP4.
• Lab Notes: Quickly polish social clips without Premiere.

OpenAI Buys io for Hardware
• Need to Know: OpenAI acquires Jony Ive’s AI device startup io for $6.5 B to build purpose‑built AI hardware.
• Lab Notes: Wow. Interesting.

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 Shannon Leonard
Publisher Privacy ∙ Publisher Terms
Substack
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture

Share