Every Flowgen release, newest first: what is new, what improved, what is fixed, and the known issues in the build you can download today.
0.1.0-beta.10: Sound, a cut, and a reason to trust the bill
The biggest build so far. Flowgen could make pictures and clips; it could not say what a run would cost before you pressed it, keep a note on why you rejected a take, put a voice under a shot, or join anything into a finished film. All of that is here. There is a Voice node and a Music & Sound node on your ElevenLabs account, an Animatic for stills on a timeline, an Edit node with a real multi-track cut, trims, transitions and a colour grade, and an export that renders exactly what the preview shows. Around them: a price and a provider balance checked before every run with a spending cap per workflow, notes and reject reasons on every take, an approved take that locks its node, a character library that holds a face across shots, a reference library, a 3D blocking node for laying out a shot before you pay for it, a brief, templates you write yourself, and a canvas header that has stopped shouting.
0.1.0-beta.9: A big workflow stops fighting back
Past a few dozen nodes the canvas went from slow to unusable. Dragging one node made Flowgen rebuild every other one, sixty times a second, and the bigger the workflow the worse it got. That work is gone. Dragging now costs the same whether you have ten nodes or two hundred. Around that are the four things people asked for once their workflows got big: an Assets panel that gathers every picture and clip the workflow has made, scrollbars so you can tell where you are, a far wider zoom range, and an Arrange that spreads your chains across the screen instead of stacking them into one endless column.
0.1.0-beta.8: Higgsfield works again
Higgsfield had stopped working completely. Not slowly, not sometimes: every price, every picture, every clip failed the moment you pressed anything, and the app could not tell you why. Higgsfield turned on a list of web addresses it will accept requests from, and a desktop app has no web address to put on it, so the browser engine inside Flowgen threw every request away before it was sent. Flowgen now makes those requests itself instead of asking the browser engine to, which is the same route it already uses for Kinovi. Also in this build: clicking your own avatar no longer blanks the window, and seventeen new models, GPT Image 2.6 and Kling 4.0 among them.
0.1.0-beta.7: Sign out
There was no way to sign out of Flowgen. There is now, in the two places people look for one: the account row at the bottom of the sidebar, which opens a real menu instead of jumping to Settings, and a block of its own in Settings. Signing out forgets your email and your Google account on this computer and asks for an address again. It leaves your workflows, your media, your provider keys and your licence exactly where they are, because those belong to the machine rather than to the address.
0.1.0-beta.6: Sign in with Google, and the Copilot on a Claude plan
Two ways in that cost nothing extra. Sign in with Google fills your name and email on the first screen, and Flowgen keeps no Google token afterwards: it asks Google one question, which address is yours, and stops. The Copilot also runs on a Claude subscription now, through Claude Code, the same way it already ran on a ChatGPT plan. Flowgen never sees the login. And every template opens for everyone, whatever you have connected, because a canvas you can see is a better reason to add a key than a greyed-out card.