Generators get pixels and a finished prompt only. Chat LLM writes the words first. Nothing from Venice/Eleven/HF language models.
Pick a track here. This sheet does not pull audio from chat.
Attach in the tray. This sheet does not pull chat history.
Upload here. Results can go to chat. Chat history does not feed this sheet.
Auto: a picture on this sheet is image-to-video. No picture is text-to-video. Both are OpenAI-compatible placeholders (Grok until you point them at your own proxy in Admin).
Image-to-video needs a file on this sheet. Output can go to chat.
Build v4.9.60
Adds this chat to the home screen (PWA). Chrome may hide the sheet if this URL was installed before.
Provider keys live in Admin. Voices and models load live. The default voice is global until you pick your own.
Download audio uses the live talk voice unless you pick another later.
TTS model is set in Admin under ElevenLabs. Voices still load live.
Verify attachments. Each type can be turned off. Audio and video are experimental. If a scan fails, the job still runs and the chat says it could not scan that file.
Which attachments
Manual Create / Edit does not ask. Auto asks once; a batch of 4 does not ask four times. Retries only run if you turn them on here or the prompt says to keep trying.
Loading limits…
No account needed. You can keep using chat as you are.
Database is SQLite. Import reads config.php into that database. MySQL is optional and does not turn on until it is connected.
Nobody signed in stays on the current limits. A signed-in person is a user. The first confirmed email becomes admin. Other roles can be added later.
Style and lyrics go to whatever music host is in config. Cover uses the same box.
T
Beta. Some tools may fail while this is in development. Say “report this” in the chat and it is saved for the owner.
@image1, @image2. Tray keeps attaches until you tap ×.You voice is what you typed. Techstar voice is the replies.
Stays in this window. Rewrite until you like it, then send. The pictures go with it.