Premiere puts Veo and Kling on the timeline, Lyria 3.5 hits the Gemini API, and Suno's fine print.
OpenAI pushed ChatGPT Images 2.5 to every ChatGPT and Codex tier on 8 September, with a Sketch tool that turns a rough drawing into a finished image and poster and merch templates. The API gets Flare, which halves latency, and Sunburst, which trades speed for finer edits. Image output still bills at $30 per million tokens, there is no batch endpoint, and Tier 1 keys get five requests a minute.
Google moved Lyria 3.5 out of Flow Music and into the Gemini app and API on 4 September, globally on web and mobile. The app adds genre and vocal pickers, templates for background beds and jingles, and a short or long track choice. The API splits into Lyria 3.5 Clip, a fixed 30 second MP3, and Lyria 3.5 Pro for full songs. Every track carries a SynthID mark, so expect labels on your video.
Believe and TuneCore will distribute Suno tracks again under a global deal announced 8 September, four months after blocking them. The fine print: it covers only Suno's unreleased industry partner model, and the current models are being retired, so nothing you have already made qualifies. The same week SOCAN sued Suno in Canada for C$20,000 per infringed work, and the caps of 20 downloads a month on Pro took effect.
Camera Intelligence's Caira app now runs Google's Gemini Omni 1.1 Flash and Runway's Aleph 2.0 on clips shot with the camera, with no text-to-video at all. Four templates, Relight, Reframe, Camera Motion and Set Extension, take plain-English tweaks; output is 720p or 1080p, takes one to three minutes, and carries a SynthID mark. The private beta opened 2 September with pricing to come in the Caira Studio tier.
One creator ran a promo video through GPT-6 Astra end to end: VRoid for the character, Blender CLI plus MCP for the classroom and motion, a four-cut 15 second storyboard as the bridge, then Omni 1.1 Flash for the anime finish. The receipts are modest: a VRM in about eight minutes on a clean pass, medium inference, and retries on faces and timing. No cost numbers, no claim of studio-ready anime.
An audit of VidMage's site found the face swap tools localised into 25 languages while the 6,107 word terms and 3,584 word privacy policy stay English everywhere; only the consent button is translated. The terms demand verifiable consent from anyone whose face you use without defining it, and quote both a 2 hour and a 24 hour deletion window. For client work, keep your own dated written releases and trust neither window.
Adobe's IBC release adds a Generative Media tool to Premiere: draw a range on any track, describe the clip or sound, and it generates in place using nearby frames as context. You pick Firefly or a partner model, including Google Veo, Kling, Runway and Luma. Sound Effects is live, Soundscape and Music are beta, and a demo reviewer heard mismatched gearbox and tyre sounds, so treat generated audio as a draft.
Microsoft put MAI-Image-2.6 and a Flash variant into public preview on Foundry on 4 September. Flash keeps the flagship's editing set, including object removal and inpainting, and by Microsoft's own numbers runs 2.8 times faster than GPT-Image-2 Medium on 72 percent less GPU. The price is the story: $19 per million image output tokens against $30 on OpenAI's new Flare and Sunburst. Flash is unranked on Arena, so test first.