COEY Cast

Alibaba’s open Z-Image model is gunning for Midjourney and Stable Diffusion Three on photoreal image generation. This episode breaks down what actually matters for teams shipping assets at scale: quality, speed, GPU requirements, and total cost per image. Learn where Z-Image shines on local workstations, how to wire it into automated thumbnail and product shot pipelines, and why prompt understandi...

Deep dive into FLUX.2 for creators and marketers who care about automation, compliance, and shipping faster. Covers what “open weight” really means, how to treat FLUX.2 dev as an R and D sandbox, and when to upgrade to commercial endpoints like pro. Breaks down production readiness signals, multi reference use cases, typography tests, and local versus hosted trade offs. Walks through policy as cod...

Google Gemini 3 is getting “I’m not going back” hype from Marc Benioff, but should teams actually switch from ChatGPT yet. This episode breaks down how to evaluate Gemini 3 and other frontier models against real workflows for creators and marketers. Learn how to design a model agnostic stack with API shims, routers, and strict schemas, what to put in a Model Rider for your contracts, and how to ru...

Grok’s Holocaust denial fallout puts AI safety in the spotlight. This episode breaks down what builders, marketers, and media operators should actually do when a model crosses the line. Learn how to design real guardrails, write honest system cards, and run multilingual red‑team testing. Get practical checklists for product, safety, marketing, and engineering teams, from kill switches and brand‑sa...

Nano Banana Pro just dropped into the Google image stack, and it is officially the mid tier workhorse between Gemini 2.5 Flash Image and Gemini 3 Pro Image. This episode digs into how to pick the right model for campaigns, how to structure prompts with schema patterns, and how to route images through critics, batching, and upscaling to cut costs. Hear real world tactics for AI FinOps, transparency...

Google’s Gemini 3, Meta’s SAM 3, and Octane AI’s CORE-1 are shaking up creator and marketer workflows. Dive deep on when to escalate to Gemini 3’s Deep Think mode for campaign planning, why SAM 3’s video segmentation is a game-changer but needs smart caching, and how CORE-1’s agentic commerce rewires ecom experiences with real guardrails. Cuts through the hype about open vs closed models, API orch...

Dive into real-world automation with Gemini 3, Veo 3 Fast, and Google’s Antigravity IDE for creators and marketers. This episode breaks down what’s mature now for AI agents and where humans still need guardrails. Get clear strategies for evaluating agent safety, integration, and brand controls at scale. Hear the latest on cost control, vendor lock-in risks, governing AI outputs, and how to wrap wo...

Alibaba’s Qwen app is opening doors for marketers in China with deep integrations and real workflow hooks, but when should global teams jump in? Grok goes free and sparks talk about the fastest ways to build real-world AI automations using tools like n8n, Make.com, and Langfuse. GPT-5.1 lands with smart heuristics for balancing speed and judgment. Hear practical KPIs for workflow automation, vendo...

Anthropic just dropped an open source framework for auditing political neutrality in language models—and it’s way more than “trust us bro.” This episode breaks down how to stress-test your AI’s even-handedness using paired prompts and structured scoring, why it matters for marketers and creators running automated content at scale, and how to bake neutrality ops into your workflows. From scoring th...

Dive into Baidu’s ERNIE 5.0—a model built to understand text, images, audio, and video as one seamless brain. Explore what “native full-modal” means, the creative perks (cross-format consistency, fewer prompt chains, and tighter factuality), and where to watch for trade-offs like latency and asset costs. We tackle ops moves like routing tasks between models, region rules, rights management, and ho...

OpenAI just dropped GPT-5.1 and creators everywhere are about to move at warp speed. Explore how the new “Instant” and “Thinking” modes in ChatGPT unlock faster first drafts and smarter, deeper workflow upgrades for marketers, content teams, and automators. Learn how to route between them, set up safeguard critics, and avoid surprise costs. Get the playbook for putting model selection, variant cap...

Google’s internal image model, aka Nano Banana 2 or GEMPIX2, has the AI grapevine buzzing but not spilling the docs. We unpack what matters for creators and marketers: cleaner text-in-image, higher resolutions, and true brand consistency could finally hit the campaign production line. Get the rundown on expected API integrations for Gemini and Vertex, the pragmatic workflow upgrades you can prep f...

AI workflows leap forward with two major China-born releases. Moonshot’s Kimi K2 Thinking brings “do the thing” automation—plan, fetch, summarize, validate, post—letting creators and marketers run multi-step projects with just enough human gatekeeping. It fits right into tool stacks, from Make to custom scripts, and can be self-hosted for privacy and control. Tencent’s open source HunyuanWorld-Mir...

Chutes is bringing open-source models to OpenAI-compatible endpoints, letting creators and marketers instantly swap in alternatives like DeepSeek‑R1 with just a base URL change. No need to rebuild your stack or decode new APIs—just plug in and roll. Get cost savings and resilience without sacrificing convenience. We break down hands-on automation recipes for content ops, copy, and creative, cover...

AI workflows just leveled up. This episode spotlights Qwen3-Max “Thinking” (closed, API only) as it shifts agentic reasoning from turn-based chat to actual planning and tool use. Hear why small vision-language models like open Qwen3-VL are making privacy, cost, and latency wins possible for image and doc triage. Explore Ming-Flash-Omni-Preview on Hugging Face bringing mixture-of-experts magic to v...

AI compute is having a Texas-sized glow-up as Microsoft secures major GPU capacity with IREN and OpenAI gears up to ride AWS’s freshest accelerators. What does that mean for creators, marketers, and operators? Get ready for faster, smoother video and content workflows, less time in queue purgatory, and more concurrency when campaigns heat up. Dive into how new hardware like NVIDIA’s GB200 and GB30...

Explore Emu3.5 from BAAI, the newest open-weight, native multimodal model that generates images and text in a single loop. Dive into how its supercharged DiDA speed claims could unlock true batch production for creators and teams, and why interleaved generation means fewer copy-paste fumbles. Learn where DIY tinkerers win, where production houses get an upgrade, and what pitfalls and possibilities...

Anthropic’s Claude Haiku 4.5 has entered the chat and it’s setting a new speed record for generative AI workflows. With blazing-fast performance, aggressive pricing, and structured outputs, Haiku is positioned as the go-to model for everyday drafting, tagging, and extracting via Anthropic’s API or Google Vertex AI. Learn how creators, marketers, and ops teams can automate titles, summaries, docs,...

NVIDIA’s open source Nemotron Nano 2 just dropped and it’s not your average small model. Packed as NIM microservices, it moves fast from “demo” to “daily workflow,” running on everything from vLLM to llama.cpp. Toggle internal reasoning with “thinking traces” for both transparency and speed. Perfect for content creators, agent devs, and ops teams who want fast, reliable, and scalable automation. E...

Apple’s Pico‑Banana‑400K is shaking up text-guided image editing by going real—not synthetic—with 400,000 photo triplets. We cover why this new dataset matters for anyone benchmarking or building image-editing automation, from marketers struggling with artifact-riddled mockups to devs craving models that finally follow instructions. Dive into its taxonomy of 35 edit operations, multi-turn edit seq...

Nvidia’s got the bottle service, but Qualcomm and Intel just crashed the party with Costco-sized savings. Qualcomm’s AI200 and AI250 and Intel’s Crescent Island promise cheaper AI inference and memory-rich accelerators that upend Nvidia’s stranglehold on creative and marketing workflows. This episode breaks down what this means for scaling LLMs, batch video variants, and TTS dubbing, plus how it i...

Alibaba has open-sourced Tongyi DeepResearch, a research agent with real web-browsing smarts and full citation trails. With code on GitHub and a friendly license, you can run it yourself and wire it into your own stack—think Slack, Notion, or even custom compliance dashboards. Under the hood, a mixture-of-experts model means you get efficiency for solo creators and scale for small teams. This epis...

Seed3D 1.0 from ByteDance takes a single image and spits out a simulation-ready 3D asset, watertight and ready for production. No downloading repos or wrangling GPUs—just upload and automate. Discover how this API-first, closed model lets ecommerce, gaming, and robotics teams scale asset creation fast. Get step-by-step workflow ideas, real-world limits around fidelity and materials, and tactical t...

Lightricks just launched LTX-2, claiming first place for open-source video models that actually sync audio and 4K visuals in a single pass. Ready for creators, marketers, and teams sick of post-production headaches. Get the scoop on native 4K, 10-second cap, multi-keyframe continuity, and creative automation that drops costs instead of frames. Hear how API access, creative studio, and open code ca...

Qwen3-VL just dropped two new vision-language models—one tiny enough to run on your phone for instant AR, brand checks, and field ops, and a 32B beast for deep enterprise automation like hours-long video review and multi-doc audits. Meanwhile, Meshy 6 Preview is making 3D asset pipelines smoother with sharper geometry and fewer cleanup headaches. This episode unpacks where these tools shine for cr...