ByteDance's Seed team put out Seedream 5.5 on 9 November. The pitch is repeat editing: change one part of an image and the untouched areas stay the same, pixel for pixel, across 100 rounds. It also outputs transparent RGBA layers, so cut-out assets need no background removal. ByteDance says severe text errors fall to 0.4%. Those figures are from ByteDance's own tests. The post gives no price, and a Pro version is still to come.
Instagram's Edits team opened Effect Builder on 9 November. You type a prompt, test the effect on your own photos and publish it to the Stories effects browser. Your handle sits on every effect, so viewers can tap through to your profile. It is free. Every effect goes through a safety review before anyone else can use it. For now it is US only and 18 and over.
DeepL's Aure model arrived on 11 November. It translates speech in real time and keeps the speaker's own voice, plus shifts in emotion, emphasis and pace. A pause before an announcement stays a pause. It works in eight languages at launch: English, German, Spanish, French, Italian, Japanese, Portuguese and Korean. It runs in DeepL Voice and the DeepL API for Voice with no setup. DeepL gives no price or latency figure, and it is built for live speech, not finished dubs.