Adobe Now Lets You Generate Soundtracks and Speech in Firefly (www.wired.com)

At Adobe MAX, Adobe doubled down on generative AI across its creative stack, with the biggest updates landing in Firefly: support for user-customizable models, a forthcoming Firefly Image Model 5 with layered image editing, and new generative audio features. Individuals can soon train custom models (Adobe says just 6–12 images to capture a character, a bit more for a “tone”) built on Firefly’s proprietary, commercially safe base; rollout begins for individuals at year-end with an early-access waitlist. Image Model 5 — expected in the months after MAX (broader availability likely in 2026) — brings native 4MP output (2K/2560×1440), prompt-based edits at 2MP or Full HD (1920×1080), and element-aware layered editing (demo showed artifact-free moving/replacing of objects like chopsticks). Adobe also introduced Generate Soundtrack (auto-suggests prompts from video with vibe/style controls) and Generate Speech (TTS using Adobe’s models plus ElevenLabs, 15 languages, and per-phrase emotion tags), plus a browser-based multi-track Firefly video editor (waiting list). Photoshop and Express get an “assistant” that mixes guidance with hands-on control, and Adobe previewed Project Moonlight (carry model/context across apps and social profiles) and an early-stage plan to surface Adobe features inside ChatGPT. Together these updates push Adobe from asset creation to an end-to-end, context-aware generative workflow with practical guardrails for commercial use.