Beyond the Frame: AI Weekly Digest #7

Welcome back to another week in AI — and wow, did the industry decide to go full throttle. Google dropped not one but two big releases with Gemini 3.1 Pro and Lyria 3, ByteDance found itself in hot water with Hollywood over Seedance 2.0, and Meta quietly patented something that'll make you think twice about your Facebook posts. On the creative side, Magnific just made AI video look a whole lot better, Qwen released a voice cloning model that only needs 3 seconds of audio, and we've rounded up the best AI film competitions running right now — including one with a $500,000 prize pool. Yeah, it's been that kind of week. Let's get into it.

Gemini 3.1 Pro Update

Google just dropped Gemini 3.1 Pro in preview, and honestly, the ".1" in the name is doing a lot of heavy lifting. This isn't a flashy, feature-packed release — it's Google quietly saying "we made it smarter." The big headline is a 77.1% score on ARC-AGI-2, which apparently means it's more than twice as good at reasoning as the original Gemini 3 Pro. It also comes with a 1 million token input window and a 65K token output limit, so if you've ever had a model cut you off mid-essay or mid-codebase, this one's got your back. You can try it now through the Gemini app (Pro and Ultra subscribers), NotebookLM, AI Studio, and even GitHub Copilot.

Lyria 3 Release in Gemini

Google's music generation model Lyria 3 just landed inside the Gemini app, and it's kind of wild. You give it a text or image prompt, and it spits out a 30-second track – complete with auto-generated lyrics, vocals, tempo control, and style options. No musical knowledge required. Every track also gets quietly watermarked with Google's SynthID tech, so AI-generated audio can be identified later. It's rolling out in beta globally to users 18 and over, and if you're a paid subscriber you get more generations per month. YouTube creators are also getting access through Dream Track, which was previously U.S.-only. It's still early days, but the ceiling here feels pretty high.

ByteDance Adds Safeguards to Seedance 2.0

So Seedance 2.0 had quite the debut. Within days of launching, the internet was flooded with AI videos of Tom Cruise fighting Brad Pitt on rooftops and Kanye West dancing through ancient Chinese palaces — all generated from text prompts in minutes. Hollywood was not amused. Disney sent a cease-and-desist, the MPA accused ByteDance of pirating copyrighted content at scale, and SAG-AFTRA jumped in over the unauthorized use of actor likenesses. ByteDance responded by promising to "strengthen safeguards," though they were pretty vague about what that actually means. The MPA has given them until February 27 to respond formally, so this one is far from over.

Magnific Video Upscaler Release

Magnific AI — the tool that became a favorite for making AI images look actually good — just launched a video upscaler, and the timing couldn't be more perfect. With Seedance 2.0 flooding the internet with 720p AI clips, Magnific is basically offering the logical next step: take those videos and push them up to 4K. The tool comes with style options like Realistic, Animation, 3D, and Artistic, plus a Turbo Mode if you're in a hurry. It's available on magnific.ai and inside the Freepik platform. If you're already in the AI video space, this one's worth bookmarking.

New Qwen Voice Cloning Model

Alibaba's Qwen team released Qwen3-TTS, and it's the kind of open-source drop that makes commercial voice cloning companies a little nervous. The model can clone a voice from just 3 seconds of audio, supports 10 languages, and apparently outperforms paid tools like ElevenLabs on some benchmarks. You can also control tone, emotion, and delivery style just by describing it in plain language — no sliders or technical settings needed. It's free, it's on Hugging Face, and it comes in multiple sizes starting from 0.6B parameters. Three seconds. Let that sink in.

Meta Gets a Patent for AI Digital Afterlife

Meta has patented a system that could keep your social media account running after you die — using an AI trained on all your posts, messages, voice notes, and reactions to respond on your behalf indefinitely. The patent, originally filed by CTO Andrew Bosworth back in 2023 and approved in late December, describes the bot handling everything from newsfeed activity to direct messages and even simulated voice or video calls. Meta was quick to clarify they have "no plans to move forward with this," but the patent exists, and Facebook is sitting on over 30 million deceased user accounts that currently generate zero ad revenue. Make of that what you will.

That's a wrap on this week's headlines — but before you close this tab, we've got something for the creators in the room. If all this AI news has you itching to actually make something, here are the best film competitions running right now where you can put those tools to work.

🎬  AI Film Competitions to Enter Right Now

AI Film Festival (AIF) — Runway

The most well-known one in the AI creator space. Films need to be 3–15 minutes long and feature generative video content. Submissions close March 31st. Winners get screened at gala events in NYC and LA, and the Grand Prix winner takes home $15,000 cash plus 1,000,000 Runway credits.

Entry fee: Free

AI for Good Film Festival — ITU/United Nations

Purpose-driven filmmaking at its finest. Deadline is March 15, 2026. Finalists get invited to the AI for Good Global Summit with an exclusive Speaker Pass and a premiere screening. Topics should tackle real-world challenges — climate, health, education, you name it.

Entry fee: Free

World AI Film Festival (WAiFF) 2026

A French festival gone full international, with the grand finale at the Palais des Festivals in Cannes. Films must use at least three generative AI tools. Hybrid live-action/AI films are welcome as long as AI plays a meaningful creative role.

Entry fee: Contact organizers at [email protected] for details

AIFFI — International AI Short Film Festival (Roatán, Honduras)

The first international AI film festival in Central America, taking place April 17–18, 2026. Categories include Best AI Grand Prix, Best Student AI Short, Best AI Animated Short, Best AI Sci-Fi Short, and Best AI Music Video. Over $10,000 USD in prizes. Deadline: March 30, 2026.

Entry fee: Varies by category — check FilmFreeway or aiffi.org

AI Shortest Film Competition — LTX Studio x Forward Festival

The ultimate low-barrier entry: films just need to be up to 60 seconds long and created with LTX Studio. Team entries of up to 3 people are welcome. Judged on creative style, originality, and storytelling. Short, scrappy, and a great way to get your feet wet.

Entry fee: Check official channels for current details

ILLUMINATE Film Festival — Micro-Short Contest

AI-assisted work is fully welcome here. Films must be under 90 seconds and submissions close April 1, 2026. Prize pool is $25,000 total — $10,000 first prize, $5,000 second, $2,500 third, and a $2,500 People's Choice Award.

Entry fee: Free via FilmFreeway

Make Your Action Scene — Higgsfield AI

This one is closing fast — only 5 days left — so if you're going to enter, now's the time. The premise is exactly what it sounds like: build a world, pick your hero, raise the stakes, and cut an action scene that makes people wish your film was real. With over 5,500 submissions already in, the competition is fierce, but so is the prize pool — $500,000 total, with $150,000 for first place, $100,000 for second, $50,000 for third, and $10,000 each for 20 Honorable Mentions.

Entry fee: Completely free to enter, and all the tools you need are built right into the Higgsfield platform.