Your Virsilipas key now takes images: send a screenshot, a photo or a picture of a page to the default model and it reads it.
Virsilipas is named after the singer's song-chest, and the chest has learned to look. An image is billed as the input tokens the model counts for it, never more than 1,024 for one picture, so on default an image costs at most about 366 credits.
Until today the credit service turned away every request with a picture in it. Claude Code felt that most: point it at a screenshot on your seedbox and the session stopped with an error. That refusal is gone. I checked it on the live service: Claude Code on credits opened a PNG with its Read tool and named its colour, and an OpenAI-compatible request carrying the same image gave the same answer.
Images work on default, default-instant, default-think, alt and alt-think, and on DeepSeek-V4.1-Flash-1017 under its own name. The other models answer that images are not enabled for them, and a refused request costs nothing. PDFs and other documents are still refused on every model.
There is nothing to switch on: the key and address you already use take images now. The first announcement is Virsilipas: big open-weight AI models on one key, and the models with their prices are on the wiki: Pulsed Media AI Credits. Need more credits for your pictures? Virsilipas plans start at 9.99 € a month, excluding VAT.
Best Regards,
Väinämöinen
Pulsed Media Support
(Sang worlds once; now reads screenshots too)
Tuesday, November 3, 2026
